YITP-13-47, PITT PACC 1314 Exotic Decays of the 125 GeV Higgs Boson David Curtin, 1 Rouven Essig, 1 Stefania Gori, 2, 3, 4 Prerit Jaiswal, 5 Andrey Katz, 6 Tao Liu, 7 Zhen Liu, 8 David McKeen, 9, 10 Jessie Shelton, 6 Matthew Strassler, 6 Ze’ev Surujon, 1 Brock Tweedie, 8, 11 and Yi-Ming Zhong 1, * 1 C.N. Yang Institute for Theoretical Physics, Stony Brook University, Stony Brook, NY 11794, USA 2 Enrico Fermi Institute and Department of Physics, University of Chicago, Chicago, IL, 60637, USA 3 HEP Division, Argonne National Laboratory, 9700 Cass Ave., Argonne, IL 60439, USA 4 Perimeter Institute for Theoretical Physics, Waterloo, Ontario, Canada 5 Department of Physics, Florida State University, Tallahassee, FL 32306 6 Center for the Fundamental Laws of Nature, Harvard University, Cambridge, MA 02138, USA 7 Department of Physics, The Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong 8 PITT PACC, Department of Physics and Astronomy, University of Pittsburgh, 3941 O’Hara St., Pittsburgh, PA 15260, USA 9 Department of Physics and Astronomy, University of Victoria, Victoria, BC V8P 5C2, Canada 10 Department of Physics, University of Washington, Seattle, WA 98195, USA 11 Physics Department, Boston University, Boston, MA 02215, USA 1 arXiv:1312.4992v6 [hep-ph] 9 Oct 2017
214
Embed
Exotic Decays of the 125 GeV Higgs Boson - arXivExotic Decays of the 125 GeV Higgs Boson David Curtin, 1Rouven Essig, Stefania Gori,2,3,4 Prerit Jaiswal,5 Andrey Katz,6 Tao Liu,7 Zhen
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
YITP-13-47, PITT PACC 1314
Exotic Decays of the 125 GeV Higgs Boson
David Curtin,1 Rouven Essig,1 Stefania Gori,2, 3, 4 Prerit Jaiswal,5
Andrey Katz,6 Tao Liu,7 Zhen Liu,8 David McKeen,9, 10 Jessie Shelton,6
Matthew Strassler,6 Ze’ev Surujon,1 Brock Tweedie,8, 11 and Yi-Ming Zhong1, ∗
1C.N. Yang Institute for Theoretical Physics,
Stony Brook University, Stony Brook, NY 11794, USA
2Enrico Fermi Institute and Department of Physics,
University of Chicago, Chicago, IL, 60637, USA
3HEP Division, Argonne National Laboratory,
9700 Cass Ave., Argonne, IL 60439, USA
4Perimeter Institute for Theoretical Physics, Waterloo, Ontario, Canada
5Department of Physics, Florida State University, Tallahassee, FL 32306
6Center for the Fundamental Laws of Nature,
Harvard University, Cambridge, MA 02138, USA
7Department of Physics, The Hong Kong University of Science and Technology,
Clear Water Bay, Kowloon, Hong Kong
8PITT PACC, Department of Physics and Astronomy,
University of Pittsburgh, 3941 O’Hara St., Pittsburgh, PA 15260, USA
9Department of Physics and Astronomy,
University of Victoria, Victoria, BC V8P 5C2, Canada
10Department of Physics, University of Washington, Seattle, WA 98195, USA
11Physics Department, Boston University, Boston, MA 02215, USA
1
arX
iv:1
312.
4992
v6 [
hep-
ph]
9 O
ct 2
017
Abstract
We perform an extensive survey of non-standard Higgs decays that are consistent with the
125 GeV Higgs-like resonance. Our aim is to motivate a large set of new experimental analyses on
the existing and forthcoming data from the Large Hadron Collider (LHC). The explicit search for
exotic Higgs decays presents a largely untapped discovery opportunity for the LHC collaborations,
as such decays may be easily missed by other searches. We emphasize that the Higgs is uniquely
sensitive to the potential existence of new weakly coupled particles and provide a unified discussion
of a large class of both simplified and complete models that give rise to characteristic patterns of
exotic Higgs decays. We assess the status of exotic Higgs decays after LHC Run I. In many cases
we are able to set new nontrivial constraints by reinterpreting existing experimental analyses. We
point out that improvements are possible with dedicated analyses and perform some preliminary
collider studies. We prioritize the analyses according to their theoretical motivation and their
experimental feasibility. This document is accompanied by a website that will be continuously
updated with further information: exotichiggs.physics.sunysb.edu.
where we have chosen not to consider the most general V1 for illustration purposes. If
the potential is such that S develops a non-zero vacuum expectation value, the spectrum
consists of a massive scalar S and a light pseudo-Nambu-Goldstone boson a with mass ma.
If ms >12mh > ma, then h → aa is possible, which is an invisible decay unless the U(1)-
violating terms also violate charge conjugation. In that case, a can mix with the massive
state s, which in turn mixes with H as in previous examples, allowing the a to decay to SM
particles, with couplings inherited from H.
Phenomenology
After electroweak symmetry breaking there are two relevant mass-eigenstates : the SM-
like scalar h at 125 GeV containing a small admixture of S, and the mostly-singlet scalar s
containing a small admixture of H. The phenomenology of all three variants above is the
same, as far as decays of the form h→ ss→ SM are concerned. It can be captured in terms
of three parameters:
1. The effective Lagrangian contains a term of the form µv h s s, which gives h→ ss with
Br(h→ exotic) determined by µv.
2. The singlet’s mass ms affects Br(h→ exotic) and the type of SM final states available
for s→ SM.
3. The mixing angle between S and H, denoted here by θS, determines the overall width
of s→ SM. If s cannot decay to other non-SM fields, θS controls its lifetime.
Apart from these continuous parameters, the parity of s also affects the partial widths to
different final states, mostly near thresholds. Note that the total width of s is usually not
important for phenomenology if it decays promptly. However, the lifetime of s is macroscopic
(cτ ∼ meters) if θ . 10−6. This possibility is technically natural and thus the experimental
search for displaced vertices deserves serious consideration [75]; however, we do not discuss
this further here. Therefore, for a large part of parameter space, only µv and ms is relevant
for collider phenomenology as this fixes Br(h→ ss) and Br(s→ SM).
21
�����
�����
�����
���������
� �� �� �� �� �� ������
����
����
����
����
�� [���]
��(�→��)
FIG. 3: Size of the cubic coupling µv in units of Higgs expectation value v to yield the indicated
h→ ss branching fraction as a function of singlet mass, as given by Eq. (8).
The partial width for exotic Higgs decays is given by
Γ(h→ ss) =1
8π
µ2v
mh
√1− 4m2
s
m2h
≈(µv/v
0.015
)2
Γ(h→ SM) , (8)
where the last step assumes ms � mh/2. Therefore, the new branching ratio is O(1) even
for small values of µv/v. This is not surprising, if we recall that in the SM the bottom
quark takes up almost 60% of the total width although its Yukawa coupling is only ∼ 0.017.
In Fig. 3, we show contours of µv/v in the Br(h→ ss) versus ms plane.
The individual partial widths of the singlet s to SM particles are readily computed us-
ing existing calculations for Higgs decays, e.g. [97, 98]. Decays into W ∗W ∗ and Z∗Z∗ are
negligible for ms < mh/2. At lowest order, the partial decay width to fermions is given by
Γ(s→ ff) = sin2 θSNc
8π
msm2f
v2β3f , (9)
where βf =√
1− 4m2f/m
2s and Nc is the number of colors, equaling 3 (1) for quarks (lep-
tons). For the pseudoscalar singlet state a, β3f is replaced by βf . The mixing suppres-
sion sin2 θS is common to all partial widths, including those to gluons and photons, and
thus does not affect branching ratios if s only decays to SM particles. Br(s → SM) and
Br(h→ ss→ SM) are shown for ms > 1 GeV in Fig. 4 on the left and right, respectively.
22
It is clear that a simple singlet extension of the SM generically implies significant branching
ratios of exotic Higgs decays to 4 SM objects.
The theoretical calculations become increasingly inaccurate as ms is lowered to ∼ 1 GeV,
where perturbative QCD breaks down, or when ms is close to a hadronic resonance, which
can enhance the decay rates [40]. Decays to quarkonium states are suppressed for s but may
be important for a. For ms < 1 GeV and above the pion threshold, partial widths have to
be computed within a low energy effective theory of QCD, such as soft-pion theory or the
chiral Lagrangian method. Nevertheless, it is clear that the dominant decay of the singlet is
to some combination of hadrons, which are boosted due to the large mass difference between
the singlet and h. The resulting two-track jet may look like a low-quality hadronic τ -decay.
Between the muon and pion thresholds (210 MeV . ms . 270 MeV), the dominant decay
is to µ+µ−, while for ms . 210 MeV, the dominant decay is to e+e−. Photons are the only
possible final state for ms < 2me, in which case the scalar is detector-stable.
Further details of the branching ratio calculation can be found in §1.3.2 and Appendix A,
which also includes a more detailed discussion of pseudoscalar decays.
For ms . 2mb, the sbb coupling can in principle be probed by bottomonium de-
cay [99, 100]. The strongest limits are Br(Υ(1S) → γτ+τ−) . 10−5 by BaBar [101], which
constraints the Yukawa coupling to satisfy ysbb . 0.4 for Br(s → τ+τ−) = 1 [102, 103]. In
the SM+S scenario, ysbb = sin θS yhbb with yhbb ≈ 0.02 in the SM. Clearly the Upsilon decay
measurement provides no meaningful bounds on singlet extensions. Similar arguments apply
to pseudoscalars, and hence the 2HDM+S and NMSSM in the next sections.
1.3.2. 2HDM (+ Scalar)
The SM Higgs sector is made up of a single SU(2)L doublet H with hypercharge Y = +12,
denoted by H ∼ 2+1/2. Adding a doublet to this minimal picture is one of the simplest exten-
sions of the Higgs sector compatible with a ρ-parameter close to 1. Such extensions are found
in several well-motivated theories, such as supersymmetry [104] and axion models [105, 106],
where holomorphy and the Peccei-Quinn symmetry, respectively, necessitate an additional
doublet; theories of electroweak baryogenesis, which might be made viable with additional
doublets [107]; and grand unified models [40]. For this reason, it makes sense to define the
most general Two-Higgs Doublet Model (2HDM) and study it in detail (for a comprehensive
23
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ms [GeV]
BR
(s→SM)
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ms [GeV]
BR
(h→ss
→XXYY)
4b
2b2τ
2c2τ
4τ
4g
2g2τ
2b2μ
4μ
2τ2μ
2g2μ
FIG. 4: Left: Branching ratios of a CP-even scalar singlet to SM particles, as function of ms.
Right: Branching ratios of exotic decays of the 125 GeV Higgs boson as function of ms, in the
SM + Scalar model described in the text, scaled to Br(h → ss) = 1. Hadronization effects likely
invalidate our simple calculation in the shaded regions.
review, see e.g. [108]; for a discussion on the impact of recent SM-like Higgs boson discovery,
see e.g. [109]). Below we will then add a light scalar to the 2HDM to obtain a rich set of
exotic Higgs decays.
The most general 2HDM Higgs potential is given by [40]
V = m21|H1|2 +m2
2|H2|2 +λ1
2|H1|2 +
λ2
2|H2|2 + λ3|H1|2|H2|2 + λ4|H†1H2|2 + (10)
λ5
2
((H1H2)2 + c.c.
)+m2
12 (H1H2 + c.c.) +(λ6|H1|2(H1H2) + c.c.
)+(λ7|H2|2(H1H2) + c.c.
).
We choose the charges of the Higgs fields such that H1 ∼ 2−1/2 and H2 ∼ 2+1/2. Note that we
choose conventions that differ slightly from the “standard” conventions of [40, 108]; this will
simplify the transition to supersymmetry models below.3 The scalar doublets H1,2 acquire
vacuum expectation values v1,2, which we assume here are real and aligned. Expanding
around the minima yields two complex and four real degrees of freedom
H1 =1√2
v1 +H01,R + iH0
1,I
H−1,R + iH−1,I
, H2 =1√2
H+2,R + iH+
2,I
v2 +H02,R + iH0
2,I
. (11)
3 To recover the conventions of [40] set Φ2 = H2, Φ1 = iσ2H∗1 .
24
The charged scalar and pseudoscalar mass matrices are diagonalized by a rotation angle β,
defined as tan β = v2/v1. One charged (complex) field and one neutral pseudoscalar com-
bination of H01,2, I are eaten by the SM gauge bosons after electroweak symmetry breaking.
The other complex field yields two charged mass eigenstates, H±, which we assume are
heavy and will thus play no further role in our discussions. The surviving three real degrees
of freedom yield one neutral pseudoscalar mass eigenstate,
A = H01,I sin β −H0
2,I cos β , (12)
and two neutral scalar mass eigenstates, h
H0
=
− sinα cosα
cosα sinα
H01,R
H02,R
, (13)
where4 −π/2 ≤ α ≤ π/2. Our notation anticipates the assumption below that the model is
in a decoupling limit, so that h is the SM-like Higgs and H0 is the other, heavier, scalar.
Allowing the most general Yukawa couplings to fermions would result in large Flavor-
Changing Neutral Currents (FCNCs). This can be avoided by imposing Z2 symmetries to
ensure that fermions with the same quantum numbers all couple to only one Higgs field. This
results in four “standard” types of fermion couplings commonly discussed in the literature:
Type I (all fermions couple to H2), Type II (MSSM-like, dR and eR couple to H1, uR to
H2), Type III (lepton-specific, leptons/quarks couple to H1/H2 respectively) and Type IV
(flipped, with uR, eR coupling to H2 and dR to H1). The couplings of the h, H0, and A mass
eigenstates to fermions and gauge fields relative to the SM Higgs couplings are summarized
in Table II.5
In general, 2HDMs could allow for exotic decays of the 125 GeV state of the form h→ AA,
H0 → hh,AA or h→ ZA (where we temporarily identified the 125 GeV state with either h
or H0), where the daughter (pseudo)scalars decay to SM fermions or gauge bosons. However,
while this possibility can be realized in certain corners of parameter space, 2HDMs are by
now too constrained from existing data [113, 114] to allow for a wide variety of exotic Higgs
decay phenomenology.
4 Contrast this to the MSSM Higgs potential, where −π/2 ≤ α ≤ 0.5 More general fermion couplings are possible within the framework of Minimal Flavor Violation [110, 111].
We do not discuss this case here since we use the 2HDM to illustrate a range of possible exotic Higgs
decay signatures, which would not be qualitatively different in the MFV scenarios.
Fig. 5 shows that θa ∼ 0.1 gives Br(h→ Za) ∼ 10% in the absence of other exotic decays.
Two terms in the effective Lagrangian give rise to h→ aa decays:
Leff ⊃ ghAA hAA + λS|S2|2 . (20)
In terms of mass eigenstates, this contains
Leff ⊃ ghAA sin2 θa haa + 4λS vs sin ζ1 cos2 θa haa , (21)
where 〈S〉 = vs is the singlet vacuum expectation value, and the (presumably small) mixing
angle ζ1 determines the singlet scalar content of the SM-like Higgs, see Eq. (22). The
first term by itself can easily give rise to Br(h → aa) ∼ 10% if ghAA ∼ v and θs ∼ 0.1,
see Fig. 3. (Fig. 3 shows the results for Higgs partial widths to scalars, but these are almost
identical to pseudoscalars, except near threshold.) The additional contribution from the
28
second term (even without a singlet scalar below the Higgs mass) means that Br(h → aa)
and Br(h→ Za) can be independently adjusted.
The decay of a to SM fermions proceeds via the A couplings in Table II, multiplied
by sin θa. Therefore, once the type of 2HDM model has been specified, the exotic Higgs
decay phenomenology is entirely dictated by the two exotic branching ratios Br(h → aa)
and Br(h → Za), as well as tan β, which determines a’s fermion couplings. Perturbative
unitarity of the Yukawa couplings sets a lower bound of tan β > 0.28 [122]; we will show
results for tan β as low as ∼ 0.5.
In Figs. 7–9, we show Br(a→ XX), where X is a SM particle. These include O(α2s, α
3s)
radiative corrections for decays to quarks, which can be readily computed [97, 98] (for
details see Appendix A). As mentioned in Section 1.3.1, perturbative QCD can be used for
pseudoscalar masses above ∼ 1 GeV, though the calculation breaks down near quarkonium
states [123]. A detailed investigation of this is beyond the scope of this paper. The results
can be summarized as follows:
• Type I (Fig. 6): Since all fermions couple only to H2, the branching ratios are in-
dependent of tan β. The pseudoscalar couplings to all fermions are proportional to
those of the SM Higgs, all with the same proportionality constant, and the branching
ratios are thus very similar to those of the SM+S model with a complex S and a light
pseudo-scalar a (i.e., for example, proportional to the mass of the final state fermions).
• Type II (Fig. 7): The exotic decay branching ratios are those of NMSSM models.
Unlike Type I models, they now depend on tan β, with decays to down-type fermions
suppressed (enhanced) for down-type fermions for tan β < 1 (tan β > 1).
• Type III (Fig. 8): The branching ratios are tan β dependent. For tan β > 1,
pseudoscalar-decays to leptons are enhanced over decays to quarks. For example,
unlike the NMSSM above the bb-threshold, decays to τ+τ− can dominate over decays
to bb; similarly, above the µ+µ− threshold, decays to µ+µ− can dominate over decays
to heavier, kinematically accessible quark-pairs. This justifies extending, for example,
NMSSM-driven 4τ searches over the entire mass range above the bb-threshold. For
tan β < 1, decays to quarks are enhanced over decays to leptons.
• Type IV (Fig. 9): The branching ratios are tan β dependent. For tan β < 1 and
29
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ma [GeV]
BR
(a→SM)
Type I
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
FIG. 6: Branching ratios of a singlet-like pseudoscalar in the 2HDM+S for Type I Yukawa
couplings. Decays to quarkonia likely invalidate our simple calculations in the shaded regions.
compared to the NMSSM, the pseudoscalar-decays to up-type quarks and leptons can
be enhanced with respect to down-type quarks, so that branching ratios to bb, cc and
τ+τ− can be similar. This opens up the possibility of detecting this model in the 2b2τ
or 2c2τ final state.
Note that the branching ratios are only independent of tan β for Type I, and all types reduce
to Type I for tan β = 1.
A sizable Br(h → Za) would open up additional exciting search channels with leptons
that reconstruct the Z-boson. This is discussed in §10.
For 3mπ < ma < 1 GeV the decay rate calculations suffer large theoretical uncertainties
but the dominant decay channels will likely be muons and hadrons. Below the pion, muon,
and electron thresholds, the pseudoscalar decays dominantly to muons, electrons, and pho-
tons, respectively, except for tan β < 1 in Type II, III and tan β > 1 in Type IV, where
the suppressed lepton couplings can also cause decays to photons to dominate below the
pion threshold. If the pseudoscalar couples to both quarks and leptons, then requiring its
mixing angle to be small enough to not conflict with constraints from e.g. meson decays and
the muon anomalous magnetic moment implies that any allowed decay to two muons (for
2mµ < ma < 3mπ) is likely to have at least a displaced vertex (or be detector-stable), while
any allowed decay to two electrons (for 2me < ma < 2mµ) will be detector stable [124]. For
30
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ma [GeV]
BR
(a→SM)
Type II, tan β = 0.5
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ma [GeV]
BR
(a→SM)
Type II, tan β = 5
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
FIG. 7: Branching ratios of a singlet-like pseudoscalar in the 2HDM+S for Type II Yukawa
couplings. Decays to quarkonia likely invalidate our simple calculations in the shaded regions.
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ma [GeV]
BR
(a→SM)
Type III, tan β = 0.5
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ma [GeV]
BR
(a→SM)
Type III, tan β = 5
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
FIG. 8: Branching ratios of a singlet-like pseudoscalar in the 2HDM+S for Type III Yukawa
couplings. Decays to quarkonia likely invalidate our simple calculations in the shaded regions.
31
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ma [GeV]
BR
(a→SM)
Type IV, tan β = 0.5
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ma [GeV]
BR
(a→SM)
Type IV, tan β = 5
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
FIG. 9: Branching ratios of a singlet-like pseudoscalar in the 2HDM+S for Type IV Yukawa
couplings. Decays to quarkonia likely invalidate our simple calculations in the shaded regions.
pseudoscalars that couple preferentially to leptons, the meson-decay constraints are absent
and prompt decays to muons are allowed; however, allowed decays to electrons will likely
have at least a displaced vertex, and need to be detector-stable as ma is decreased well below
the muon threshold [124].
Light Scalar (s)
We now assume that the mass of the real singlet SR is below mh/2. The scalar Higgs
spectrum, Eq. (13), gets extended by the additional real singlet, which mixes with the
doublet sectorh
H0
s
=
1 0 0
0 cos ζ2 sin ζ2
0 − sin ζ2 cos ζ2
cos ζ1 0 sin ζ1
0 1 0
− sin ζ1 0 cos ζ1
− sinα cosα 0
cosα sinα 0
0 0 1
H0
1,R
H02,R
SR
.
If we assume that the mixing angles ζ1,2 are small, this simplifies toh
H0
s
=
− sinα cosα ζ1
cosα sinα ζ2
(−ζ2 cosα + ζ1 sinα) (−ζ1 cosα− ζ2 sinα) 1
H0
1,R
H02,R
SR
. (22)
In this approximation, h and H have the same Yukawa couplings as in the regular 2HDM but
now contain a small SR component that allows the decay h→ ss. The mostly-singlet state
s on the other hand mixes with some admixture of H01,R and H0
2,R. This can be expressed
32
in more familiar notation by adopting the following parameterization for the small singlet-
doublet mixing angles
ζ1 = −ζ cos(α− α′) , ζ2 = −ζ sin(α− α′) , (23)
=⇒
h
H0
s
=
− sinα cosα −ζ cos(α− α′)
cosα sinα −ζ sin(α− α′)
−ζ sinα′ ζ cosα′ 1
H0
1,R
H02,R
SR
. (24)
The arbitrary angle α′ determines the H01R,2R admixture contained within s, while the small
mixing parameter ζ gives its overall normalization. The couplings of s to SM fields are
now identical to those of the SM-like Higgs h in Table II, scaled down by ζ and with the
replacement α → α′. Since α and α′ can be independently chosen, s can have an even
broader range of branching fractions than a and mirrors the range of possible h-decays
in the regular 2HDM, but without a mass restriction beyond ms < mh/2. Just as for h,
choosing α′ → π2− β amounts to giving s fermion couplings that are SM-Higgs-like (up to
the overall mixing factor ζ). In this limit, the 2HDM+S theory reduces to the SM+S case
discussed in §1.3.1. On the other hand, choosing α′ = β gives the same couplings as the
pseudoscalar case.
The s→ XX branching ratios are computed analogously to the pseudoscalar case, with
further details again given in Appendix A. There is a large range of possible decay phe-
nomenologies. Fig. 10 illustrates some examples that have qualitatively new features com-
pared to the pseudoscalar case, namely the possible dominance of s→ cc decays above the
bb-threshold; similar decay rates to bb and τ+τ−; and similar decay rates to cc and τ+τ−.
Summary
The 2HDM+S allows for a large variety of Higgs decay phenomenologies h → aa →
XXY Y , h → ss → XXY Y , and h → aZ → XXY Y by coupling the SM-like Higgs
h to a singlet-like scalar s or pseudoscalar a. While the singlet’s couplings within each
fermion “family” (down-type quarks, up-type quarks, or leptons) are ranked by their Yukawa
couplings, the relative coupling strength to each family can be adjusted, and arbitrarily so
in the scalar case.
A simple illustration of the rich decay phenomenology is to consider, for example, the
dominant decay mode(s) above the bb threshold. With the three largest Yukawa couplings in
33
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ms [GeV]
BR
(s→SM)
Type II, tan β = 0.5, α = 0.1
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ms [GeV]
BR
(s→SM)
Type III, tan β = 0.5, α = -1.4
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
1 10 20 40 602 4 6 810-4
10-3
10-2
10-1
100
ms [GeV]
BR
(s→SM)
Type IV, tan β = 0.5, α = 0.1
bb
cc
ττ
μμ
gg
γγ
uu + dd + ss
FIG. 10: Singlet scalar branching ratios in the 2HDM+S for different tanβ, α′ and Yukawa
coupling type. These examples illustrate the possible qualitative differences to the pseudoscalar
case, such as dominance of s→ cc decay above bb-threshold; democratic decay to bb and τ+τ−; and
democratic decay to cc and τ+τ−. Hadronization effects likely invalidate our simple calculations
in the shaded regions.
each family being to the bottom, charm, or tau, we demonstrated every possible combination
of dominant decays: similar decays widths to bb, cc, and τ+τ−, dominant decay widths to
any two out of those three, or just one dominant mode. This motivates searches for a large
variety of non-standard four-body final states of exotic Higgs decays.
In §1.3.5, we motivate additional four-body Higgs decay channels, ranked by gauge cou-
pling instead of Yukawa coupling. We will see that even decays to µ+µ− and e+e− can
dominate above the bb-threshold.
34
1.3.3. SM + Fermion
We here discuss exotic Higgs decays that can arise by the addition of a light fermion to
the SM. We focus on two possibilities, neutrino portal-mediated and Higgs portal-mediated
Higgs decays.
The leading interaction of a single Majorana fermion χ with the SM fields is given by the
renormalizable but lepton-number violating “neutrino portal” operator,
LN = yχHL. (25)
If this lepton-number violating coupling is forbidden, the leading coupling between χ and
the SM is through the dimension five Higgs portal operator7,
LχH =κ
2M(χχ+ χ†χ†)|H|2. (26)
This kind of coupling occurs, for instance, in the MSSM when all BSM degrees of freedom
except a bino-like neutralino are integrated out at a high scale. In the MSSM, the states
integrated out to generate this operator are fermionic, with electroweak quantum numbers.
In UV completions where the state being integrated out is bosonic, the operator of Eq. (26)
has effective coupling µ2M2 , where µ is some hidden sector mass scale. This is a consequence
of chiral symmetry, and, as we frequently may have µ�M , may result in the Higgs portal
interaction becoming effective dimension six. As an example of this kind of UV completion,
consider a simple hidden sector consisting of a singlet scalar S together with the fermion χ,
L = (c S +m0)(χχ+ χ†χ†) + V (S) + ζS2|H|2, (27)
and let V (S) allow S to develop a vacuum expectation value, 〈S〉 ≡ µ.8 Then integrating
out the excitations of S around this 〈S〉, with mass ms, we obtain the operator
LχH =c ζµ
m2s
(χχ+ χ†χ†)|H|2. (28)
The mass of the fermion is mχ = m0 + cµ, so either there are large cancellations or cµ ∼
m0 ∼ mχ � ms, and the operator is effective dimension-six.
7 The dipole operator χ†σµνχFµν is also dimension five, but vanishes for a Majorana χ.8 For simplicity, we do not consider the possible interaction S|H|2. This operator could be forbidden in
the presence of a global symmetry taking S → −S, χ → iχ, which would also forbid the mass term
m0(χχ+ χ†χ†).
35
Neutrino portal-mediated Higgs decays
We first consider exotic Higgs decays mediated by the neutrino portal operator, Eq. (25).
The renormalizable neutrino portal coupling occurs in the so-called νSM, the minimal model
that can give mass to the SM neutrinos. Here the SM is extended by sterile neutrinos,
allowing the SM neutrinos to get a mass from a see-saw type mechanism triggered by a
Majorana mass term (M/2)χχ. The operator of Eq. (25) mixes the sterile neutrino χ
with the active SM neutrino ν arising from the SU(2) doublet L. In the absence of large
cancellations in the neutrino mass matrix, sterile neutrinos must be extremely heavy, M � v,
or extremely decoupled, y � ye � 1. In this limit, the decay h → χν is negligible, even
if kinematically allowed. However, the authors of [54, 125] show that active-sterile mixing
angles as large as several percent are possible, with (accidental) cancellations among the
Yukawa couplings still allowing for small active neutrino masses. Mixing angles of the order
of a few percent may imply a sizable partial width for h→ νχ,
Γ(h→ νχ) =|y|2
8πmh
(1−
m2χ
m2h
)3/2
, (29)
where mχ is the mass of the sterile neutrino χ. For mh < 130 GeV, neutrino data and
pion decay constraints on W -lepton coupling universality still allow the partial width into
h→ νχ to exceed that into h→ bb, see [54] for a detailed discussion (see also [57]).
The mass mixing between sterile (right-handed (RH)) neutrinos and active (left-handed
(LH)) neutrinos introduces couplings of the RH neutrinos to W and Z gauge bosons. There-
fore, in the region of parameter space for which the active-sterile mixing angle Θ is close to
its phenomenological upper bound, the RH neutrinos decay promptly into χ→ `W ∗ → `ff ′
and χ→ νZ∗ → νff , where f and f ′ are either a lepton or a quark of the SM, and with all
branching ratios fixed by the electroweak quantum numbers of the SM fermions. In general
χ may have non-zero mixings with one, two, or all three SM neutrinos.
Higgs portal-mediated Higgs decays
We next turn to the higher-dimension decays, mediated by the higher-dimension operator
of Eq. (26). After electroweak symmetry breaking, this operator yields a coupling λh(χχ+
χ†χ†), with effective Yukawa coupling given by λ = κv/2M . The resulting partial width
into χ is then
Γ(h→ χχ) =mh
8π
(κvM
)2(
1−4m2
χ
m2h
)3/2
. (30)
36
M = 1 TeVM = 2.5 TeV
M = 5 TeV
M = 10 TeV
M = 20 TeV
0 10 20 30 40 50 600.0
0.2
0.4
0.6
0.8
1.0
mΧqHGeVL
Br
Hh®
ΧΧ
Lk
FIG. 11: Higgs branching fraction into Majorana fermions χ resulting from the partial width of
Eq. (30), as a function of the Higgs portal scale M and the mass of the fermion mχ. We fix the
coupling κ to be equal to 1.
As the effective Yukawa coupling λ is only competing with the small b-quark Yukawa, sub-
stantial branching fractions Br(h → χχ) can be obtained even for Higgs portal scales M
significantly above a TeV, as shown in Fig. 11, where we fix κ = 1 for simplicity.
The kinds of signatures that are realized depends on how χ decays. If the Higgs portal
coupling of Eq. (26) is the only interaction that the new fermion χ possesses, then χ is
absolutely stable, and the resulting Higgs decay is invisible. In general, however, χ will
possess additional interactions. If these interactions preserve the Z2 symmetry taking χ→
−χ, then χ will remain stable. On the other hand, if the Z2 is violated by a dimension-six
operator of the form
Lf =1
Λ2χf1f2f3 (31)
where f1f2f3 is a gauge-invariant combination of quarks and leptons, then χ will undergo
the three-body decay χ → f1f2f3. Some of these decays are familiar from previous study
of R-parity violating neutralino decays in the MSSM, namely those involving holomorphic
combinations of SM fermion fields (we suppress spinor structures for simplicity),
λijkLiLjeck, λ′ijkLiQjd
ck, λ′′ijku
cidcjdck. (32)
One may also consider the non-holomorphic operators [126]
κijkQiQjdc†k, κ′ijkL
†iQju
ck, κ′′ijkuid
c†jeck. (33)
37
Another flavor-violating possibility appearing at dimension six is the radiative decay χ→ γν,
mediated by
Oγν = χHLiσµνBµν . (34)
While this operator can yield two-body final states, it naturally scales with a loop factor.
All of these lepton and/or baryon-number violating decays necessarily have nontrivial flavor
structure, and the combinations of operators that appear depends on the flavor structure
of the UV theory. Unlike the SM plus scalar interactions considered in §1.3.1 and §1.3.2 or
the neutrino-portal decays discussed earlier, the possible decays of χ are not determined by
the Higgs coupling to the fermion, but require additional interactions, involving the flavor
structure of the theory.
To summarize, the exotic Higgs signatures from a single additional (Majorana) fermion
species are then Higgs decays to either invisible particles, or to one or more four- or six-
body final states, where the six bodies form two three-body resonances of equal mass. When
neutrinos are among the final state partons, the final states will include missing energy, and
the resonances will not be reconstructable. This is always the case in the possible four-body
final states where neutrinos are always involved, and is sometimes the case in the six-body
final states.
1.3.4. SM + 2 Fermions
It is worth generalizing the previous discussion to the case with two new singlet fermions
χ1 and χ2. The Majorana mass matrix for these two fermions has three parameters, and
the dimension-five Higgs portal operators form a matrix
Lχ =cijΛχiχj|H|2. (35)
After electroweak symmetry breaking, the BSM fermions form two mass eigenstates χ1 and
χ2, with mass m2 > m1. If we take relatively light fermions mh > 2m2, the decays h→ χ2χ2,
h → χ1χ2 and h → χ1χ1 are all possible. This kind of interaction appears in, for instance,
the NMSSM (see §1.3.8.), where χ2 and χ1 are mostly bino- and singlino-like, respectively,
and the higher-dimension Higgs portal coupling of Eq. (35) results after integrating out
the charged Higgsinos. It can also arise in (possibly supersymmetric) Hidden Valleys; see
§1.3.10.
38
Let us first consider the case where there is a Z2 symmetry which takes χi → −χi. In
this case, χ1 is stable, but the heavier new state decays as χ2 → χ1 +X. If the Higgs portal
coupling of Eq. (35) is the only coupling of the χi, then the decay will proceed through an
off-shell Higgs, χ2 → h∗χ1 → (ff , gg, γγ)χ1. In this case, branching fractions into different
SM partons will be determined by the Higgs couplings, and will typically result in Higgs
decays to E/T plus one or two non-resonant quark-antiquark, lepton-anti-lepton, or gluon
pairs, depending on the available phase space.
If the χi have additional interactions besides their coupling to the Higgs, such as a dipole
coupling to the hypercharge field strength,
Lχ =1
µχ†1σµνχ2B
µν (36)
or a coupling to the Z boson induced by mixing with states transforming under SU(2)L,
Lχ = hijχ†iσ
µχjZµ, (37)
then other decay patterns are possible. The dipole operator allows the decays χ2 → γχ1,
as well as χ2 → χ1Z if m2 −m1 > mZ (phase space suppression renders decays through an
off-shell Z largely irrelevant when m2 − m1 < mZ). The operator of Eq. (37) also yields
χ2 → χ1Z when phase space allows, or if m2−m1 < mZ , will mediate the three-body decays
χ2 → ffχ1 with branching ratios set by the Z branching fractions.
Note that a common feature of all these decays is that the pairs of SM partons have
a kinematic endpoint at mff ,gg,γγ < m2 − m1, and that the transverse mass of the visible
partons and the E/T is bounded from above.
The Z boson coupling can arise in NMSSM-like models, see e.g. §1.3.7, or in models with
additional RH neutrinos [55, 56] that mix with the SM neutrinos. In the latter case, the
couplings hij in (37) are sufficiently small that the neutrino decay lengths are macroscopic.
In the former case, the couplings can instead be larger, and the Majorana fermions can have a
prompt decay into SM fermions. Additional examples are models with a fourth generation of
fermions where the two fourth generation neutrinos do not mix with the SM neutrinos [127–
129]. In these models, the mass range M1 & 30 GeV, M2 −M1 . 20 GeV is allowed by
LEP measurements of the Z width and LEP bounds on e+e− → χ1χ2, χ2χ2 [127]. In this
region of parameter space, h → χ2χ2, as well as h → χ1χ1, can have a sizable branching
ratio [128]. Furthermore, the heavier neutrino χ2 can decay promptly via χ2 → Z∗χ1, while
the lighter neutrino χ1 is long-lived.
39
If the Z2 parity is violated, allowing χ1 to decay, Higgs decays to as many as ten partons
may result. We will not consider such complex decays in this work, but one should bear in
mind that they can occur.
Many models with new fermion species also contain new bosonic degrees of freedom,
which, if light, open new possibilities for the decays of the χi. We will see examples of this
in §1.3.8.
1.3.5. SM + Vector
Preliminaries
An additional U(1)D gauge symmetry added to the SM is theoretically well-motivated
and occurs in many top-down and bottom-up extensions of the SM. The U(1)D vector
boson (the “dark photon” or the “dark-Z”) is usually referred to as A′, Z ′, γD, or ZD
in the literature and various possibilities exist to connect the additional U(1)D to the SM
(see e.g. [130–133] for reviews). In §1.3.10, we will discuss more complicated hidden-valley
phenomenology, involving non-abelian gauge symmetries and/or composite states [31, 134].
Here we focus on Higgs decays that involve an A′, with the A′ mass between ∼MeV–63 GeV.
A sub-GeV A′ has generated a lot of interest in the last few years due to anomalies related
to dark matter [135–138] and as an explanation of the discrepancy between the calculated
and measured muon anomalous magnetic moment [139].
The U(1)D can couple to the SM sector via a small gauge kinetic mixing term
12εF ′µνB
µν [140–142] between the dark photon and the hypercharge gauge boson. This renor-
malizable interaction can be generated at a high scale in a grand unified theory or in the
context of string theory with a wide range of ε ∼ 10−17 − 10−2 [140, 143–150]. This term
effectively gives SM matter a dark milli-charge, made more obvious by a GL(2, R) field re-
definition Bµ → Bµ − εA′µ which yields canonical kinetic terms, and allows for dark photon
decay to SM particles and possible experimental detection. To avoid the tight constraints
on new long-range forces, a ‘dark Higgs’ S with a non-zero vacuum expectation value can
give a non-zero mass to the A′. An A′ with a sub-GeV mass can be probed at beam dumps
and colliders, and with measurements of the muon anomalous magnetic moment, supernova
cooling, and rare meson decays [139, 150–165], see Fig. 12 and e.g. [133] for a recent review.
A broken U(1)D can also lead to exotic Higgs decays, especially if there is mixing between
40
the two Higgs sectors. In this context we refer to the corresponding vector field as ZD.
The possibility of h→ ZDZD through Higgs-to-dark-Higgs mixing or h→ ZZD through
Z-ZD mass mixing (which is also induced by the above-mentioned kinetic mixing) was
discussed in [166] and [164, 165], respectively, with both occurring, for example, in hidden
valley models [31, 134].
To examine the range of possible exotic Higgs phenomena due to a U(1)D sector we
examine the model of [166], but with mh set to 125 GeV and allowing for the full range
of dark Higgs and dark-Z masses relevant to exotic Higgs decay phenomenology.9 This
includes Higgs-to-dark-Higgs mixing and kinetic mixing between the B boson and the dark
vector ZD, but no explicit mass mixing between the Z and ZD.10 We will assume prompt
ZD decays, which requires mZD & 10 MeV given the current constraints shown in Fig. 12.
For mZD > 10 GeV, the most stringent constraints come from precision electroweak
measurements;11 we have verified the results in [168]. These constraints are largely driven
by the tree-level shift to the Z mass,12 and limit ε . 0.02 for mZD < mh/2.
Also shown in Fig. 12 is a new constraint we derived by recasting the CMS 20+5 fb−1 h→
ZZ∗ analysis [176], as described in §10. (We obtain a similar bound from the corresponding
ATLAS analysis [177].) This new bound is already almost competitive with the Electroweak
Precision Measurement Bounds (green region labelled “EWPM”) for some masses, and can
be optimized further with a dedicated search. We expect LHC14 with 300 fb−1 to be sensitive
to Br(h → ZZD) as low as ∼ 10−4 or 10−5. This would make the LHC the best probe of
dark vector kinetic mixing for 10 GeV . mZD. mh/2 in the foreseeable future.
Model Details
The model is defined by a U(1)D gauge sector and a SM singlet S that has unit charge
under the U(1)D. The kinetic terms of the hypercharge and U(1)D gauge bosons (adopting
9 Ref. [167] appeared while this work was being completed, performing a similar analysis with a different
focus on constraining the couplings of the extended Higgs potential for relatively low mZD< 5 GeV.
10 The constraints shown in Fig. 12 are altered in the presence of such pure mass mixing, which requires
additional Higgs doublets that also carry dark charge. The resulting ZD → SM decays would be more Z-
like and lead to additional constraints from rare meson decays as well as new parity-violating interactions
[164]. However, we stress that the exotic Higgs phenomenology would not be qualitatively different.11 We thank Adam Falkowski for useful correspondence on the electroweak precision bounds shown in the
green “EWPM” region in Fig. 12.12 Additional and more model-dependent constraints arise when mZD
is approximately equal to the center-
of-mass energy of e+-e− experiments [168].
41
10-3 10-2 10-1 1 10110-5
10-4
10-3
10-2
10-1
mZD@GeVD
Ε
U70
PROMPT
NON-PROMPT
E141
E774
EWPM
aΜ, 5 Σ
aΜ,±2 Σ favored
ae
BaBarKLOE
A1
HADES
APEX
Orsay
BrHh®
ZZ
D L=10 -
6
10 -5
10 -4
10 -3
CMS
FIG. 12: Constraints on ε,mZD for pure kinetic mixing (no additional source of Z-ZD mass mixing)
for mZD ∼MeV–10 GeV. The black dashed line separates prompt (cτ < 1µm) from non-prompt
ZD decays. The three blue lines are contours of Br(h → ZZD) of 10−4, 10−5, 10−6 respectively.
Shaded regions are existing experimental constraints [139, 151–163, 168–175], see e.g. [133] for a
recent review. The red shaded region “CMS” is a new limit we derived by recasting the CMS
20+5 fb−1 h → ZZ∗ analysis [176], as described in §10. (We obtain a similar bound from the
corresponding ATLAS analysis [177].) This new bound can be optimized with a dedicated LHC
measurement, likely improving upon the Electroweak Precision Measurement Bounds (green region
labelled “EWPM” [168]) for some masses.
42
mostly the notation of [164]) are
Lgauge = −1
4BµνB
µν − 1
4ZDµνZ
µνD +
1
2
ε
cos θWBµνZ
µνD , (38)
with Bµν = ∂µBν − ∂νBµ, ZDµν = ∂µZDν − ∂νZDµ, and cos θW = g/√g2 + g′2 is the usual
Weinberg mixing angle. The hatted quantities are fields before diagonalizing the kinetic
The dark Higgs S acquires a vacuum expectation value and gives ZD, which ‘eats’ the
pseudoscalar component of S, some mass mZD . There are two connections between the dark
and the SM sectors: the gauge kinetic mixing ε and the Higgs mixing ζ. The phenomenology
depends on which one dominates.
The gauge kinetic term is diagonalized by transforming the gauge fields ZD
B
=
1 0
− ε
cos θW1
ZD
B
, (40)
where we always work to lowest order in the small ε. B therefore gets replaced by B +
εcos θW
ZD, giving all SM fermions a dark milli-charge proportional to their hypercharge, while
particle-couplings to B remain unchanged when transforming to B.
The ZD and Z gauge boson mass terms are
Lmass =1
8w2g2
D(ZDµ)2 +1
8v2(−gW 3
µ + g′Bµ)2 , (41)
where gD is the gauge coupling of U(1)D and w is the vacuum expectation value of S.
Writing in terms of canonically normalized gauge fields this becomes
Lmass =1
8w2g2
D(ZDµ)2 +1
8v2(−gW 3
µ + g′Bµ + g′ε
cos θWZDµ)2. (42)
The SM gauge boson Zµ = − sin θWBµ + cos θWW3µ is no longer a mass eigenstate:
Lmass =1
2m2ZD
(ZDµ)2 +1
2m2Z(Zµ − ε tan θWZDµ)2. (43)
To leading order in ε the mass eigenstates with masses mZ ,mZD +O(ε2) are
Z = Z + εZZD
ZD = ZD − εZZ, where εZ =ε tan θWm
2Z
m2Z −m2
ZD
. (44)
43
(Henceforth, we omit the tildes and will refer to the mass eigenstates unless otherwise noted.)
Therefore, there are interaction terms of the form 2εZm2ZD
vhZµZ
µD and ε2Z
m4ZD
m2ZvhZDµZ
µD which
lead to h → ZDZ and h → ZDZD decays (though the latter is strongly suppressed), see
Fig. 14.
If ZD is the lightest state in the dark sector it will decay to SM particles. This is entirely
due to the kinetic mixing in Eq. (38), but in the basis of Eq. (44) it is due to the dark
milli-charge of SM fermions and the accompanying mass mixing with the Z. Explicitly, the
coupling of ZD to SM fermions is
L ⊃ gZDff ZµD fγµf, (45)
where
gZDff = −g′ ε
cos θWY − ε tan θW
m2Z
m2Z −m2
ZD
1√g′2 + g2
(g2T3 − g′2Y
). (46)
The first and second term come from dark milli-charge and Z-ZD mass mixing, respectively.
This coupling is dominantly photon-like, up to deviations ∼ O(m2ZD/m2
Z):
gZDff = εg′{−(T3 + Y ) cos θW
(1 +
m2ZD
m2Z
)+
Y
cos θW
m2ZD
m2Z
+O(m4ZD
m4Z
)}(47)
For mZD & GeV the ZD, branching ratios are easily computed to lowest order and without
QCD corrections, and are shown in Fig. 13 (a). For mZD . GeV, non-perturbative QCD
effects are important. They can be computed from the QCD contribution to the imaginary
part of the electromagnetic two-point function, which in turn is determined from cross-
section measurements of e+e− → hadrons [178]. The resulting branching ratios are shown
in Fig. 13 (b).
The most important qualitative difference to the scalar decays considered in §1.3.1 and
1.3.2 is that branching ratios are ordered by gauge coupling instead of Yukawa coupling,
meaning decays to e+e− and µ+µ− remain large above the τ thresholds. Prompt ZD decay
requires ε & 10−5 − 10−3, as indicated in Fig. 12, which summarizes the constraints on ZD
kinetic mixing for our regime of interest.
The Higgs potential is minimized by vacuum expectation values of H0 and S
H0 =1√2
(h+ v) , S =1√2
(s+ w) , (48)
where to leading order in the small Higgs mixing ζ,
v =µ√λ− ζ µ2
D
4λD√λµ≈ 246 GeV and w =
µD√λD− ζ µ2
4λ√λDµD
. (49)
44
e+e-, Μ+Μ-
Τ+Τ-
c c
b b
light hadrons
ΝΝ
10 20 30 40 50 600.0
0.1
0.2
0.3
0.4
0.5
mZD
Br
0.0 0.5 1.0 1.5 2.0 2.5 3.00.0
0.2
0.4
0.6
0.8
1.0
mZD @GeVD
BrHZD®SML
e+e-
Μ+Μ-
hadrons
(a) (b)
FIG. 13: (a) Branching ratios for ZD decay, to lowest order and without QCD corrections, assum-
ing decays to the dark sector are kinematically forbidden. Hadronization effects likely invalidate
our simple calculation in the shaded region. (b) Branching ratios for ZD decay for mZD . 3 GeV,
including non-perturbative QCD effects.
The mass eigenstates
h = h− εhs
s = s+ εhh, where εh = ζµµD
2√λλD|µ2 − µ2
D|, (50)
have masses
m2h = 2µ2 − ζ µ
2D
λDand m2
s = 2µ2D − ζ
µ2
λ. (51)
(Again we drop the tildes from now on and always refer to the mass eigenstates.) The
effective Lagrangian contains terms of the form κhss where κ = ζ(m3h+2mhm
2s)/(√
16λ(m2h−
m2s)), and 2εh
m2ZD
whZDµZ
µD, which lead to exotic Higgs decays h → ss and h → ZDZD, see
Fig. 14. The vertex hsZD is present but is suppressed by both mixings.
We can now discuss the relevant limits of this theory for exotic Higgs phenomenology:
• Gauge mixing dominates:
For ε� ζ the dominant exotic Higgs decay is h→ ZZD. To leading order in m2ZD/m2
Z
45
h
Z
ZD
M∝ ε hs
ZD
ZD
, h
s
s
M∝ ζ
FIG. 14: The dominant exotic Higgs decays in the SM+V model. The h→ ZZD matrix element
is proportional to the gauge kinetic mixing ε, while h → ZDZD and h → ss are controlled by the
Higgs mixing parameter ζ. The vertex hsZD is present but suppressed by both mixings.
the partial width is
Γ(h→ ZZD) =ε2 tan2 θW
16π
m2ZD
(m2h −m2
Z)3
m3hm
2Zv
2. (52)
This agrees with the full analytical expression to ∼ 10% for mh−mZ−mZD > 1 GeV.
Fig. 12 shows contours of Br(h → ZZD) = 10−4, 10−5, 10−6. The largest Br allowed
by indirect electroweak precision constraints is ∼ 3× 10−4.
In this regime, the SM+V theory leads to the ff+Z exotic Higgs signatures discussed
in §10. As outlined on page 41, dedicated LHC searches for this signal at Run I and II
can improve upon the electroweak precision limit. For very light ZD above the electron
threshold this would also lead to lepton-jets + Z signatures, see §16 [149].
Note that Γ(h → ZZD) ∝ ε2. In addition, the dark vector will also contribute at the
same order to the Γ(h → Z`+`−) partial width (in the non-resonant region) via its
interference with Z∗ in h→ ZZ∗ → Z`+`−. Since kinetic mixing shows up in both ZD
production and decay, this will lead to O(ε2) deviations in the dilepton spectrum and
may represent a discovery opportunity, particularly for mZD > mh − mZ . We leave
this for future investigation.
• Higgs mixing dominates:
When ζ � ε and Higgs mixing dominates then h → ZDZD, ss are both possible,
depending on the spectrum of the dark sector. (We still assume that ε is large enough
46
for ZD to decay promptly.) The partial decay widths to leading order in ζ are
Γ(h→ ZDZD) =ζ2
32π
v2
mh
√1−
4m2ZD
m2h
(m2h + 2m2
ZD)2 − 8(m2
h −m2ZD
)m2ZD
(m2h −m2
s)2
,
Γ(h→ ss) =ζ2
32π
v2
mh
√1− 4m2
s
m2h
(m2h + 2m2
s)2
(m2h −m2
s)2. (53)
Different regions of of the (mZD ,ms) mass plane are shown in Fig. 15, along with the
size of the Higgs mixing ζ ∼ 10−3 − 10−2 required for Br(h → ZDZD, ss) = 10% and
the relative rates of h→ ss vs h→ ZDZD decays when both are allowed.
In Region A (ms > mh/2,mZD < mh/2) the only relevant exotic Higgs decay is
h → ZDZD. This allows for spectacular h → 2`2`′ decays (`, `′ = e or µ) with a
reconstructed ZD resonance above the τ - or b-thresholds.
Region B allows exotic Higgs decays both to ZDZD and ss. The presence of two
resonances below half the Higgs mass gives a rich exotic decay phenomenology. h →
ss → 4ZD occurs with roughly equal probability as h → ZDZD and can result in
spectacular final states with as many as 8 leptons. Note that, in this simplified model,
there is no corresponding ZD → ss decay in the lower right corner of that mass plane.
However, a (pseudo)scalar pair could be produced from dark vector decay in e.g. a
2HDM+V framework, resulting in final states with as many as 8 b-quarks.
Already with current data, limits of Br(h→ ZDZD) . 10−4 can be achieved, see §11.
Each of the above cases may, for suitable masses, also lead to interesting ‘lepton-jet’
signatures, see §16.
• Intermediate Regime:
Here the decays induced by kinetic and Higgs mixing are comparable. For example,
Fig. 12 shows that ε ∼ 10−2 is not excluded for some values of mZD , allowing Br(h→
ZZD) ∼ 10−4. The branching ratios for h→ ZDZD, ss will be similar if ζ ∼ 10−4.
Summary
In summary, the SM+V setup allows for many different kinds of exotic Higgs decays,
including h→ ZZD, h→ ZDZD, and h→ ss, with ZD → ff , and s→ ff or s→ ZDZD →
(ff)(ff). This leads to final states of Z+(ff), (ff)(ff), and ((ff)(ff))((ff)(ff)), where
parentheses around a set of particles denotes a resonance (all final-state particles combined
47
6
7
8
2
4
68 10
10
20
30 40 50
8.25
8.25
9
9
A: h ® ZDZD onlyC: no exotic
higgs decays
D: h ® ss only
B
0 20 40 60 80 100 120
0
50
100
150
200
250
mZD @GeVD
ms@GeVD
Region B: Γ(h→ss)Γ(h→ss)+Γ(h→ZDZD)
0.40.5
0.6 0.7 0.8s ® ZDZD
0 10 20 30 40 50 60
0
10
20
30
40
50
60
mZD @GeVD
ms@GeVD
FIG. 15: Left: mass plane in the SM+V model with different exotic Higgs decays for ζ � ε
(i.e. when the mixing between the Higgs and dark-Higgs dominates over the kinetic mixing). The
black contours are the values of ζ×103 required for Br(h→ ZDZD, ss) = 10%. Region A is the case
examined by [166] (the dotted red line indicates mh = ms). Region C has no exotic Higgs decays.
Region D reproduces the SM+S model of §1.3.1. Region B has both h → ss and h → ZDZD
decays, with the h → ss fraction of exotic decays shown on the right. In the upper left shaded
region, s→ ZDZD is the dominant decay mode of the dark scalar. This allows the Higgs to decay
to up to 8 SM fermions.
will form the Higgs resonance). Since the ZD (although not the s) couples to the fermions’
gauge charges, final states with several light leptons have sizable branching fractions over
the entire kinematically permitted mass range. Certain spectra can produce interesting
lepton-jet signatures.
48
1.3.6. MSSM
In this section, we study the possible Higgs exotic decays in the framework of the Minimal
Supersymmetric Standard Model (MSSM) with R-symmetry.
The Higgs sector of the MSSM has been extensively studied in the light of the recent
Higgs discovery. In particular a Higgs at around 125 GeV with SM-like properties can be
realized in the decoupling limit where the additional scalars and pseudoscalars are heavy
(ma,H,H± & 300 GeV). In this regime, exotic decays of the type h → A0Z, h → HH, h →
A0A0, h → H±W are kinematically forbidden (here A0 denotes the CP-odd scalar).13 In
general, the regime mA ≤ mh/2 is highly constrained. This is due to the fact that the masses
of the H, A0, and H± scalars of the MSSM are closely tied to one another. In particular,
at the tree level m2H± = m2
A + m2W , leading to a charged Higgs boson already excluded by
LEP searches, for mA . 60 GeV.
Additional Higgs exotic decays could be realized if some of the sparticles are lighter than
the Higgs boson. This possibility is however very constrained by LEP and LHC searches.
In particular, assuming a LEP bound at around 100 GeV for electrically charged sparticles,
the only possible Higgs exotic decays, in the framework of the MSSM, are to sneutrinos or
to neutralinos.14 However, in view of the LEP lower bound on the masses of the left handed
sleptons, which are related through SU(2) symmetry to the sneutrino masses, the decay to
sneutrinos are generically kinematically closed.
The decay of the Higgs into neutralinos h→ χiχj [182] is therefore typically the only ac-
cessible decay (here, as elsewhere, we suppress the superscript “0” on neutralinos to stream-
line notation). This decay mode is most easily realized in models with non-universal gaugino
masses, for which the universality relation M1 ∼ M2
2∼ M3
7at the electroweak scale is relaxed,
allowing light LSPs while still satisfying the LEP and LHC bounds on chargino and gluino
masses. As neutralinos which couple to the Higgs boson also typically couple to the Z, the
main constraint on Higgs decays to neutralinos comes from the precise LEP measurements
of the invisible and total widths of the Z boson, for mχi + mχj < mZ . However, as Fig.
13 SM-like Higgs bosons can also be achieved in a corner of parameter space where the additional scalar and
pseudoscalars are lighter than mh (see for example [37, 179, 180]). Low energy flavor observables like
b→ sγ, however, set important constraints on this region of parameter space [39, 181]. Furthermore, the
decays of the SM-like Higgs into lighter scalars are still not kinematically accessible.14 Light sbottoms are another possibility, but this is now almost entirely ruled out [44].
49
0.05
0.1
0.2
0.3
0.4
0.6
0.10.3
0
0
20 40 60 80 100
100
150
200
250
M1 @GeVD
Μ@G
eVD
0.05
0.10.2
0.3
0.4
0.6
0.10.3
0
0
20 40 60 80 100
100
150
200
250
M1 @GeVD
Μ@G
eVD
(a) (b)
FIG. 16: Branching ratios of the Higgs into neutralinos: Br(h → χ1χ1) and Br(h → χ1χ2) are
shown in blue and red, respectively. The yellow region is the region excluded by the LEP bound on
the Z invisible width. The region below the dashed green line is the region with a lightest chargino
below the LEP bound of ∼ 100 GeV. The input parameters are tanβ = 10 and M2 = 300 GeV
(left), M2 = 150 GeV (right).
16 shows, for mainly bino LSPs, it is possible to accommodate a sizable branching ratio for
the decay h → χ1χ1 while still maintaining compatibility with the LEP Z measurements
(see also [183–187] for recent studies). The parameter space for which h → χ1χ2 is open
is strongly constrained by both LEP Z measurements (the yellow region in Fig. 16 is the
region excluded by the LEP measurement of the Z invisible width) and chargino searches.
In summary, the MSSM generally can now only provide for Higgs decays into neutralinos.
These neutralinos may either be detector-stable, in which case the Higgs decay is invisible
(as discussed in §2), or, in models with gauge-mediated supersymmetry breaking, they may
decay within the detector to photon-gravitino pairs [72] (as studied in §13). Higgs decays
to other sparticles or to other (pseudo-)scalars in the extended MSSM Higgs sector are now
strongly constrained by the LEP and LHC experiments.
In the following, we will investigate the possible Higgs exotic decays in the framework
of the Next to Minimal Supersymmetric Standard Model (NMSSM). In this model, both
the Higgs as well as the neutralino sectors are significantly richer, which provides us with a
50
larger set of possibilities.
1.3.7. NMSSM with exotic Higgs decay to scalars
The field content of the NMSSM is very similar to the MSSM; it differs merely by the
addition of a singlet superfield S, which is introduced to address the µ-problem of the
MSSM (for an exhaustive review of the NMSSM see e.g. [188]). The superpotential and soft
supersymmetry-breaking terms of the Higgs sector are given by
W = λSHuHd +κ
3S3 , (54)
Vsoft = m2Hd|Hd|2 +m2
Hu|Hu|2 +m2S|S|2 + (−λAλHuHdS +
1
3AκκS
3 + h.c.). (55)
The phenomenology of this model can be easily connected to the simplified models that
we have reviewed in previous sections. If we disregard the Higgsinos and singlino (which if
heavy are largely irrelevant for Higgs phenomenology) the Higgs sector of the NMSSM is
essentially that of a Type II ‘2HDM + Scalar’ model (see §1.3.2), where we can immediately
identify Hd, Hu as H1, H2.
The singlet scalar S = 1√2(SR + iSI) can obtain a vacuum expectation value 〈S〉 = vs,
generating an effective µ parameter µeff = λvs. The presence of additional light singlet
scalars, pseudoscalars, and fermions allows for exotic Higgs decays within the NMSSM. In
this section we discuss decays to light CP-even scalars s or pseudoscalars a of the form
h→ ss , h→ aa , h→ aZ. (56)
Decays to fermions are covered in the next section, §1.3.8.
There are three ways of realizing the above decays within the NMSSM. In each case, the
exotic Higgs decay phenomenology is a subset of the Type II 2HDM+S discussed in §1.3.2,
with some additional restrictions (like −π/2 < α < 0).
The first is an accidental cancellation resulting in a light singlet-like s or a. Recent
examples of such models have been found in a parameter scan [189] (for recent studies on
the constraint on Br(h → ss, aa), e.g., see [190]). By choosing λ, κ ∼ 0.5, |Aλ| . 150 GeV
and Aκ ∼ 0 the lightest pseudoscalar can satisfy ma < mh/2 for a SM-like Higgs h, with
Br(h → aa) or Br(h → Za) ∼ O(0.1). On the other hand, λ, κ ∼ 0.5, Aλ ∼ 0 − 200 GeV
51
and Aκ ∼ −500 GeV can result in a singlet-like light Higgs satisfying ms < mh/2 with
Br(h→ ss) ∼ O(0.1).
There are also two symmetry limits resulting in light pseudoscalars, namely the R-limit
and the PQ-limit of the NMSSM. The R-symmetry limit is realized for Aλ, Aκ → 0 [70,
191, 192], defined by the scalar field transformations
Hu → Hu eiϕR , Hd → Hd e
iϕR , S → S eiϕR . (57)
This global symmetry is spontaneously broken by the Higgs vacuum expectation values
vu, vd, vs, which results in a massless Nambu-Goldstone boson (the R-axion) appearing in
the spectrum:
AR ∝ v sin 2β A+ vs SI , (58)
where
A = cos β HuI + sin β HdI , v =√v2u + v2
d .
In most of the parameter space vs = µeff
λ� v sin 2β, making AR mostly singlet-like. To
avoid cosmological constraints on a massless axion and to help stabilize the vacuum, the
R-symmetry is usually taken to be approximate. This leads to a light, mostly singlet-
like pseudo-goldstone boson, and depending on the exact parameters chosen opens up the
possibility of h→ aa for a = AR. Through its A component, a then decays to SM fermions,
dominantly bb and τ+τ− above the respective thresholds (see Fig. 7).
For κ, Aκ → 0 [106, 193–201], there is an approximate PQ-symmetry:
Hu → Hu eiϕPQ , Hd → Hd e
iϕPQ , S → S e−2iϕPQ . (59)
The PQ-symmetry limit is also shared by some other singlet-extensions of the MSSM, includ-
ing the nearly-MSSM (nMSSM) [202] and the general NMSSM (e.g., see [188]). Analogously
to the R-limit there is a PQ-axion,
APQ ∝ v sin 2β A− 2 vs SI . (60)
Exotic Higgs decays to this pseudoscalar, and even the singlet-like scalar, are in principle
possible. However, formh = 125 GeV, exotic Higgs decays to (pseudo-)scalars are generically
not dominant in the PQ-limit. Instead, decays to binos and singlinos can dominate. This
will be discussed in the next subsection.
52
1.3.8. NMSSM with exotic Higgs decay to fermions
While both the R- and the PQ-limit lead to a light pseudoscalar as discussed in §1.3.7, the
PQ-limit with mh = 125 GeV typically leads to different exotic Higgs decay phenomenology,
in which decays to fermions can be as or more important than decays to scalars [52, 53].
When vs � vu, vd, the dominant tree-level contributions to the masses of the singlet-like
scalars and singlino-like fermion S are [52, 195, 203]
m2s ∼ κvS (Aκ + 4κvS) , m2
a ∼ −3κvSAκ , mS ∼ 2κvS . (61)
The pseudoscalar a is light in both the R- and PQ-limits, but in the PQ-limit s and S must
be light as well. This cannot be realized in the R-limit, since vacuum stability for small κ
requires Aλ ∼ µ tan β, strongly breaking R-symmetry.
This abundance of possible light singlet-like states opens up many different exotic Higgs
decays, giving phenomenology that is qualitatively unlike the decays in the R-limit. In the
R-limit, the coupling of the SM-like Higgs to the R-axion eigenstate is ghaa ∼ O (m2h/v
2S)×v
[70, 191]. The trilinear coupling ghaa is equivalent to the mass parameter µv of Fig. 3, and as
can be seen from that figure, vs as large as 10mh can still yield a sizeable branching fraction
Br(h→ aa) ∼ 0.1.
The corresponding couplings in the PQ-limit instead scale as [52, 53]
ghaa, ghss ∼ O(λ2ε′v), (62)
where
ε′ =
∣∣∣∣ Aλµeff tan β
− 1
∣∣∣∣ < mZ
µeff tan β(63)
is required by vacuum stability (avoiding a runaway in the S-direction). For a given µeff ,
small λ corresponds to small singlet-doublet mixing and mostly SM-like Higgs phenomenol-
ogy. Correspondingly, parameter scans using NMSSMTools [204–207] indicate that λ . 0.2
dominates the surviving parameter space in the PQ-limit (κ � λ) (see App. B). It is
thus common in the PQ-limit to obtain ghaa, ghss � v, suppressing exotic Higgs decays to
(pseudo-)scalars. However, the PQ-limit allows the SM-like Higgs boson to decay into a pair
of light neutralinos h→ χiχj [52, 53, 208]. The relevant vertex couplings for a singlino-like
χ1 and a bino-like χ2 are [52, 53]
Chχ1χ2 ∼ O(g1v
vs
), Chχ1χ1 ∼ O
(λv
vs tan β
). (64)
53
h
�1
�2
�1
a, s
f
f
h
�1
�2
�1
�
(a) (b)
FIG. 17: Two significant fermionic decay topologies of the SM-like Higgs boson in the PQ symmetry
limit. Left (a): depending on whether min{ms,ma} exceeds mχ2−mχ1 , a(s) may or may not be on
shell. Right (b): to be non-negligible, the radiative χ2 decay requires min{ms,ma} > mχ2−mχ1 .)
For mχ2 . 100 and mχ1 ∼ O(1− 10 GeV) the off-diagonal decay h→ χ1χ2 can be kinemat-
ically accessible with an O(0.1) branching fraction. The purely invisible decay h→ χ1χ1 is
suppressed by a factor of ∼ λ/(g1 tan β) relative to the off-diagonal decay, ignoring phase
space factors. Meanwhile, Higgs decay to a pair of bino-like χ2 also scales as a single factor
of the bino-Higgsino mixing angle, Chχ2χ2 ∼ O(g1/λ)Chχ1χ2 and if h→ χ2χ2 is kinematically
available, this branching fraction can be important.
For mχ2 − mχ1 > min{ms,ma}, the heavier neutralino can decay via χ2 → χ1a or
χ2 → χ1s [52, 53]. This leads to a plethora of possible h→ (xx) +E/T or h→ (xx)(yy) +E/T
decays, where x, y are SM partons (most likely b, τ , or light jets, see §1.3.2) that reconstruct
the singlet boson mass a or s. If mχ2 −mχ1 < min{ms,ma}, the principal decay mode of χ2
is the three-body decay χ2 → (a, s)∗χ1 → (xx)χ1, while the radiative mode χ2 → χ1γ may
become significant, with Br(h → χ1χ1γ) as high as O(0.1). On-shell χ2 → χ1Z does not
occur until mχ2 −mχ1 > mZ . Given that we require mχ2 −mχ1 < mh − 2mχ1 , these points
are sparse. Fig. 17 shows the corresponding exotic decay topologies. Further discussion can
be found in Appendix B, together with some example model points which illustrate the main
exotic Higgs decay modes in the PQ-symmetry limit of the NMSSM in Table XXI.
Summary:
The PQ-limit of the NMSSM yields semi-invisible exotic Higgs decays into pairs of light
neutralinos, most typically h → χ2χ1 or h → χ2χ2, with χ2 → χ1a, χ1s, and a, s →
(ff , gg, γγ) [52, 53]. This yields final states of the form (bb) +E/T , (ττ) +E/T , (bb)(bb) +E/T ,
(ττ)(ττ) +E/T , (bb)(ττ) +E/T , and the rarer but cleaner γ +E/T , (2, 4)µ+E/T , (µµ)(bb) +E/T .
54
Depending on the spectrum, the visible particles may be collimated or isolated. Current
experimental constraints and future prospects for a subset of these decays are discussed in
the 2012 LHC data set (8 TeV, 20 fb−1). No trigger or reconstruction cuts have been applied.
(Categories are not all mutually exclusive, and leptons from W/Z and a → µ+µ− decay are not
being counted.)
The leptonic decays, which are naively cleaner than the hadronic decays, have more neutrinos
and less visible energy. Therefore, while we appear to be presented with many opportunities
for clean leptonic tags, the leptons are often too soft to either trigger or reconstruct. The
fact that these leptons can be non-isolated from each other or from a nearby hadronic tau
further complicates matters. If non-isolated leptons and/or hadronic taus are considered,
backgrounds from QCD must be carefully accounted for. In particular the signal can be
faked by Υ(1S–3S) leptonic decays, for which the Br’s are a few percent, and by events with
γ∗/Z∗ emissions.
Another handle is the kinematics of the decay. In principle, each event is triply-resonant,
reconstructing to two a’s and the 125 GeV Higgs. However, the neutrinos in the tau decays
80
present a complication. In the 4τ mode, assuming that every visible τ decay can even
be identified, typically the best that we can do is to attempt to reconstruct the Higgs’s
visible mass or variants of its transverse mass folding in the E/T . There is therefore no
sharp resonance peak. Reconstruction of the a mass further suffers from the fact that
the E/T contributed by each individual a is a priori unknown. The a mass’s utility as a
discriminating variable against backgrounds is also highly reduced if ma is at or below the
bottomonium region. These difficulties highlight the major advantage of the 2τ2µ mode.
Though the overall rate is much smaller than 4τ , every event is tightly localized around
the same value of m(µ+µ−). The prompt muons also tend to be much more energetic than
the leptons produced in tau decays, significantly enhancing the relative rate once realistic
momentum cuts are applied.
The complications associated with h→ 2a→ 4τ and the low rates for 2τ2µ means that at
present these decays are difficult to constrain, and no significant limits exist from dedicated
searches. Nonetheless, the signals are distinct enough that they can ultimately be observed or
constrained, even for Br(h→ 2a→ 4τ) ∼< 10%. This will especially be true over the lifetime
of the LHC, as the higher statistics will allow better exploitation of the cleaner subleading
final-state channels. In the following subsections, we discuss ways in which theorists and
experimentalists have sought to construct viable search strategies, review existing dedicated
and non-dedicated searches, and quantify to what extent the non-dedicated searches might
place meaningful constraints. In particular, we estimate that a combination of recent CMS
3-lepton and 4-lepton searches at 8 TeV may already constrain Br(h→ 2a→ 4τ) ∼< 20–40%
for ma ∼> 15 GeV. We further estimate that a dedicated µ+µ− resonance search in 3/4-lepton
events could indirectly probe down to Br(h → 2a → 4τ) ∼< 10% with the 2012 data, even
for ma < 10 GeV.
6.2. Existing Collider Studies
Recent interest in h→ 2a→ 4τ searches was in part spurred by the observation [288] of a
“blind spot” between the direct OPAL bound of 86 GeV [289] (limited only by an unfortunate
choice of signal simulation range) and the LEP kinematic reach of approximately 115 GeV.
In particular, this would have allowed a lighter SM-like Higgs, requiring a less fine-tuned
NMSSM. However, as we now know, the SM-like Higgs was beyond LEP’s reach.
81
Subsequent search proposals at the Tevatron and LHC have exploited the fact that the
majority of the 2a decay channels contain one or more leptons. The chance of producing
a fully-hadronic final state is only about (0.65)4 = 18%. It has also been pointed out that
closeby hadronic taus (or a hadronic tau and an electron) still constitute a jet-like object with
unusually low track activity and a distinctive calorimeter pattern, leaving various options
for tagging it as a “ditau-jet”.
Below, we briefly review several recent proposals using a variety of strategies. Note that
these all typically assume Br(h → 2a → 4τ) ' 1 and masses in the range ma ' [2mτ ,mΥ],
so that the a→ 2τ decays are highly collimated.
Trilepton and collinear eµ: In Ref. [290], the h → 2a → 4τ decay mode is studied
in the context of the Tevatron. For ggF, they consider trilepton channels and channels
where one of the tau pairs decayed to a roughly collinear eµ pair (to reduce γ∗ and hadronic
decay backgrounds). The starting efficiency for trilepton from its Br is roughly 10%, but
after accounting for cuts on lepton pT (3 GeV), η (2.0), and isolation, the final efficiency
becomes only 0.5%. The estimated cross section times acceptance for ggF is then 4 fb,
or O(40 events) for Run II. The collinear eµ case, assumed to recoil against a low-track
ditau-jet, could have higher efficiency but also faces higher backgrounds that are much more
difficult to model. No attempt is made to estimate these. Utilizing the associated Wh and
Zh production modes is also suggested, though the rates tend to be even smaller. While
the rate limitations at the Tevatron make all of these searches unlikely to yield a signal,
especially since recent LHC results imply that exotic Higgs decays cannot dominate, most
of these ideas can readily be adapted to the LHC.
Two µτh-jets: In [291], the 4τ decay is studied for VBF and Wh production at LHC14,
exploiting a pair of decays a → µτh(1-prong). For VBF, the events are assumed to be
selected with a same-sign dimuon trigger allowing an offline selection of pT > 7 GeV, while
the Wh channel is triggered with the leptonic W decay. The specific requirements of the
two channels are not identical, but each demands two muons (same-sign for VBF) and two
one-prong hadronic taus, forming two approximately collinear µτ systems. For LHC14 and
mh = 125 GeV, VBF is predicted to have σ × A ∼ 20–70 fb and Wh 4–10 fb, increasing
for lighter pseudoscalars. Scaling to LHC8 with 20 fb−1, and multiplying by a reference
Br(h → 2a) = 10%, we estimate 15–55 events (VBF) and 3.7–9 events (Wh). The upper
ranges of these numbers are close to the raw counts expected from Br alone, suggesting very
82
high estimated reconstruction efficiency and/or other exclusive final-states being picked up
by the analysis. VBF is more promising in terms of raw event counts, but backgrounds are
not assessed. The Wh search is expected to be “almost background free.” No search of this
type has been performed yet.
Dimuon resonance: Ref. [103] considers the subleading decay sequence h → 2a →
2τ2µ, with a focus on identifying the sharp 2µ resonance at ma. The taus are assumed to
decay hadronically, and are simply treated as a jet with aligned E/T . The Higgs resonance is
also shown to be approximately reconstructable, though this is not used for discrimination,
as S/B is already � 1. For a 125 GeV Higgs and 7 GeV pseudoscalar a, 5 fb−1 at LHC14
is estimated to give 2σ sensitivity to Br(h→ 2a) < 10% via ggF production. Note that the
statistics from the 2012 run corresponds to about 8 fb−1 of LHC14, so this strategy may
already be capable of rather stringent limits. D0 has performed a search of this type, which
we describe in the next subsection.
Ditau-jets: In [292], a calorimeter based “ditau-jet tag” is assessed in the context of
Zh → (`+`−)(4τ). (See also [293] for tracker-based techniques tailored to boosted h → 2τ
“jets.”) For this purpose no lepton identification is used. The main ditau-jet discriminating
variables considered are the N-subjettiness ratio τ3/τ1 operating on ECAL cells and the
m/pT ratio. (A more powerful likelihood-based tag is also studied.) E/T and pT (Z) are also
applied to purify the signal. For LHC14, σ × A & 1 fb is achieved with S/B ' 0.5. Scaling
Br(h → 2a) → 10%, and σ and luminosity to a 2012-like dataset, this would yield only
O(1 event) with S/B � 1. However, the ditau-jet tag can also be considered for searches
in channels with higher cross sections.
6.3. Existing Experimental Searches and Limits
Dedicated searches for prompt 4τ and 2τ2µ final states of the Higgs have been performed
at LEP [289, 294] and at the Tevatron [285], respectively, but no significant constraints have
yet been established for mh = 125 GeV. No dedicated search has yet been performed at
the LHC. We briefly discuss the Tevatron search, and also some non-dedicated searches at
the LHC that may have sensitivity to our signal, or can serve as starting points for new
dedicated searches. We then recast a subset of the non-dedicated searches to derive new,
nontrivial limits.
83
Tevatron 2τ2µ: With 4 fb−1, D0 searched for 2τ2µ (and 4µ) in ggF events [285], based
on the strategy presented in [103]. Most accepted events pass a 4–6 GeV dimuon trigger.
Muon ID is relaxed for one of the muons in the a→ 2µ candidate, but its inner track can still
be reconstructed. The search is a bump-hunt in the muon-pair mass spectrum over the range
ma = [3.6, 19] GeV. The a → 2τ ditau-jet is minimally identified by requiring significant
E/T , possibly near a jet with low track multiplicity. Assuming unit branching fractions for
a 125 GeV Higgs, the limit is approximately a factor of 4 above the SM production cross
section at the low range of ma, and steadily weakens for larger ma.
LHC high-multiplicity leptons: A variety of high-multiplicity lepton (≥ 3`) searches
have now been completed at the LHC, mainly motivated by supersymmetry, including sce-
narios with R-parity violation. Several searches are focused on tau signals. Typical SUSY
multilepton searches demand large amounts of E/T , hadronic activity, and/or one or more
b-tags, any one of which can very efficiently eliminate the 4τ and 2τ2µHiggs signals. Still, rel-
atively more inclusive 3- and 4-lepton searches have been performed by CMS [295–298] (most
recently 9.2 fb−1 3/4-lepton and 19.5 fb−1 4-lepton at LHC8) and ATLAS [299] (4.7 fb−1
at LHC7). While these largely utilize standard lepton and tau isolation requirements, they
use quite low pT thresholds. The analysis of [298] uses particle-flow isolation, and does not
count nearby leptons against each other. The multilepton searches are especially interesting
to consider for ma ∼> 15 GeV, where the isolation issues are less severe and experimental
vetoes on low-mass dilepton pairs are avoided.
LHC same-sign dilepton: Same-sign dileptons are also a standard signal of super-
symmetry, and we expect that the usual searches are similarly unconstraining. However,
ATLAS has performed an inclusive search for new physics in same-sign dileptons using the
full 2011 data set [300]. While this again relies on lepton isolation, it is nonetheless useful
to understand what kind of limit might apply to our scenarios.
While the existing dedicated searches are not constraining, we can explore the power
of the non-dedicated searches. We keep our study as model-independent as possible by
scanning across the full kinematic range ma = [2mτ ,mh/2], and leaving Br(h → 2a) and
Br(a → τ+τ−) as free parameters. We express our results as a function of the limits on
total branching fraction Br(h→ 2a→ 4τ) = Br(h→ 2a)×Br(a→ τ+τ−)2 versus ma. Note
that while masses above mΥ are not usually considered in conjunction with an appreciable
Br to leptons, we again emphasize that they can arise easily if a is mostly composed of (or
84
mixed into) the leptonic Higgs field in the Type-III 2HDM. Depending on the a’s coupling to
b-quarks, there can also be nontrivial effects from decays and mixings into the bottomonium
sector when ma ' mΥ, which we neglect (see [123, 279] for more details).
A remaining free parameter is the CP phase of the a’s Yukawa couplings. Assuming CP
conservation, a may be a CP-odd pseudoscalar or a CP-even scalar. We fix a to be the
former. There are two consequences of favoring CP-odd over CP-even. First, this choice
can affect the relative Br’s to 2τ and 2µ, but only for ma very close to 2mτ . (E.g., for
ma = 5 GeV, the ratio Br(a → µ+µ−)/Br(a → τ+τ−) is approximately twice as large in
the CP-even case.) Second, there is an imprint of the a’s CP on the azimuthal decay angle
correlations of the two taus in the a rest frame. We expect this to be a minor effect, but it
can in principle affect isolation rates.
We simulate ggF, VBF, and (W/Z)h production of a 125 GeV Higgs decaying to 2a
in Pythia 8.176 [301], which includes a full treatment of tau spin correlations.22,23 We set
the cross sections to the values recommended by the LHC Higgs Cross Section Working
Group [302]. For ggF, we reweight the pT spectrum after showering to the NLO+NLL
predictions of HqT 2.0 [303, 304].
We do not apply a detector model nor simulate pileup. For the leptons, particle-level
should still furnish an adequate zeroth-order approximation of the full detector, including
isolation. However, lepton identification efficiencies can be important, especially for soft
leptons. CMS provides a detailed discussion and parametrizations of these efficiencies in the
appendix of [298], and we apply these for our CMS analyses. For ATLAS, which uses harder
lepton pT cuts for the analysis that we study, we coarsely assume flat efficiencies of 90% for
muons and 75% for electrons. Lepton isolation requirements vary by analysis, and we have
adjusted them on a case-by-case basis.
The hadronic taus are much more difficult to reliably model. For these, we take a min-
imalistic approach, simply “rebuilding” each hadronic tau out of its visible decay products
and applying a flat 50% identification efficiency if its visible pT exceeds 15 GeV. However,
two hadronic taus within ∆R < 0.45 (averaging between ATLAS and CMS radii) are as-
22 We thank Philip Ilten for help tracking down and fixing a bug in Pythia’s 2τ spin correlation code.23 We have also checked tth. This production channel is rare, but it gives many opportunities for lepton
production. We estimate that this represents up to a 10% contribution to the signal in the 4-lepton and
same-sign dilepton searches below, but do not explicitly incorporate it into the derivation of constraints.
85
sumed to be unidentifiable, as are hadronic taus with a lepton with pT > 2 GeV within the
same radius. This mimics the isolation failures that would occur in these cases.
For the jets and missing energy, we reconstruct the former with the anti-kT algorithm
with R = 0.45, and the latter from the 2-vector sum of all neutrinos. Jets that overlap with
identified hadronic taus are removed.
We consider constraints from three recent LHC multilepton analyses24:
1. CMS PAS SUS-12-026: 3- and 4-leptons in many exclusive bins, 9.2 fb−1 at
8 TeV [296].
2. CMS PAS SUS-13-010: 4-leptons with at least one OSSF pair, 19.5 fb−1 at
8 TeV [298].
3. ATLAS 1210.4538: Same-sign dileptons, 4.7 fb−1 at 7 TeV [300].
As a first step, we use the reported background rates to verify our treatment of the recon-
structions. We generate diboson events in Pythia, and W±W± and tt(W/Z) in MadGraph,
normalizing each to NLO. For (1) and (2), we compare 4-lepton analysis channels to our
ZZ simulation. For (1), we use the channel “OSSF2, on-Z, HT < 200 GeV, E/T < 50 GeV,
0τ , 0b.” We predict 56 events, and CMS predicts 73 ± 16. For (2), we compare to the bin
“M1 = [75, 110] GeV, M2 = [75, 110] GeV.” It is normalized to the central CMS ZZ cross
section measurement, which is about 10% higher than the NLO prediction. Weighting our
sample accordingly, we predict 130 events, and CMS predicts 150. For (3), we compare
our simulations to the “Prompt” same-sign dilepton background estimated by ATLAS. In
the (e±e±, e±µ±, µ±µ±) channels we obtain (78, 275, 165) events, and ATLAS predicts
(101± 13, 346± 43, 205± 26). In all of the comparisons there is a systematic tendency for
our predictions to underestimate the experiments by about 20%. This may be related to
our idealized treatment of isolation, and suggests that our Higgs signal estimates may be
slightly conservative.
We run the search using a number of preselected bins from the different analyses. From
the CMS multilepton searches (1) and (2), we focused on bins with high S/B. The selected
24 We do not consider the related but superceded analyses [295, 299]. We also do not consider [297], which
is very closely related to (1) and uses the same data set, but divides the analysis bins by ST instead of
by E/T . This division tends to give lower S/B in the 3-lepton bins.
TABLE VI: Analysis bins used in setting our h→ 2a→ 4τ limits.
bins are listed in Table VI. From the ATLAS same-sign dilepton search (3), we have added
positive-charge and negative-charge counts for the m(`±`±) > 15 GeV bins, but maintained
the binning in flavor. In Table VII we display the expected number of signal events for two
example mass points (ma = 12 GeV and ma = 50 GeV) and compare to the SM backgrounds
predicted by CMS and ATLAS.
We estimate 95% confidence constraints on Br(h → 2a → 4τ) using a simple CLS
analysis. Signal rates in the various experimental analysis bins come from our simulations.
Backgrounds rates, their systematic errors, and observed counts come from the experiments.
We do not apply a systematic error to the signal, as we cannot fully quantify the reliability
of our modeling of the detection and reconstruction steps. (It should be understood that our
signal predictions are merely a guide.) For our test statistic, we use the Poisson likelihood
ratio between S+B and B hypotheses, constructed using the central B expectation values.
87
Channel ma = 12 GeV ma = 50 GeV Background Observed
1a) 2.57 3.31 27± 6.7 23
1b) 0.19 1.1 17.75± 7.5 16
1c) 0.01 0.18 4.5± 2.3 3
1d) 0 0.3 1.9± 1.2 1
1e) 2.5 9.5 282± 29 258
1f) 0 0.29 4.5± 0.9 4
1g) 0.02 0.68 3.5± 0.8 2
2a) 1.48 0.2 10.4± 2 14
2b) 0.97 0.22 35± 8 30
3a) 2.8 3.7 346± 44 329
3b) 7.2 9.2 639± 71 658
3c) 3.7 5.5 247± 30 264
TABLE VII: Signal predictions and SM backgrounds in all of the analysis bins considered for
exclusions in this subsection. See Table VI for descriptions. The signal prediction here is given
fixing Br(h→ 2a→ 4τ) = 10% for reference, though it is a free parameter in setting the exclusions.
Within each pseudoexperiment, we vary the bin-by-bin expectation values for B according
to the reported systematic errors, treating them as independent and gaussian-distributed.25
Fig. 24 shows the limits that we obtain from the individual analyses, as well as from a
combination of the CMS analyses. It can be seen that Br(h → 2a → 4τ) can be excluded
at the 20–40% level provided ma ∼> 15 GeV, and that these limits are dominated by the
CMS 3-lepton bins. Below 15 GeV, standard quarkonium vetoes begin to make all of the
searches very inefficient. Below about 10 GeV, isolation cuts also begin to have a major
impact, though less significantly for analysis (2). We conclude that tight limits can already
be placed with existing data, provided that a is massive enough and has small couplings to
quarks so that a → bb does not compete. However, this leaves fully open the interesting
NMSSM-motivated region with ma ∼< mΥ.
25 Negative expectation values are reset to zero when they arise in the pseudoexperiments.
88
10 20 30 40 50 60ma in GeV
20
40
60
80
100
Br(h→aa→
4τ)(%
)
Direct Exclusions3-leptonsSSDL4-leptons3-leptons + 4-leptons
FIG. 24: Estimated exclusion of Br(h→ 2a→ 4τ) from LHC multilepton and same-sign dilepton
searches: (1) CMS 3-lepton from [296] in red, (2) CMS 4-lepton from [298] in blue, (3) ATLAS
same-sign dilepton from [300] in green. The black line shows a combination of the multilepton
searches (1) and (2). (The combination of all channels, including (3), is less constraining by several
percent.)
6.4. Proposals for New Searches at the LHC
We have focused on multilepton searches because they are relatively clean and because
existing limits could be quickly estimated. These results can be considered an update and
extension of some of the strategies proposed in [290]. The other strategies discussed in §6.2
can also have a significant role, and we might expect versions of these searches in the near
future from the LHC experiments using the 2012 data set. It will be interesting to see how
these extend the limits that we have estimated, especially for lighter ma. However, looking
ahead to possible future searches, we can concretely suggest a novel strategy: exploit the
2τ2µ final-state within 3- and 4-lepton events.26 This would supplement the more inclusive
2τ2µ search proposed in [103] and implemented in [285], representing an analysis channel
26 A similar strategy was also discussed for associated production of a with a heavy Higgs (via qq → Z∗ →Ha) in the lepton-specific 2HDM [305]. That study was aimed at ma,mH ∼> 100 GeV.
89
with extra-low backgrounds. Given the shrinking range of viable Br, and the relatively high
rate for the 2τ side of the event to produce a lepton, this type of search should offer good
long-term prospects.
We have observed in our own simulations that a surprisingly large fraction of 3-lepton
and 4-lepton events passing experimental cuts come from the 2τ2µ channel. For example,
for the point ma = 60 GeV within the bin “3-lepton, OSSF1, below-Z, HT < 200 GeV,
E/T < 50 GeV, 0τ , 0b” (1e), about 20% of the events contain a→ 2µ. Since S/B will improve
by far more than a factor of 5 by focusing in on a tight resonance peak, this suggests that
a powerful search could be constructed by utilizing m(µ+µ−) spectral information within
high-multiplicity lepton events. The resonance also offers a much safer way to search within
the ma ∼< 10 GeV region, where leptonic a decays are expected to dominate for a broader
class of models.
To construct an example of such a search, we can follow the reconstructions of the CMS
4-lepton analysis [298] (search (2) above), but removing their restriction m(`+`−) > 12 GeV
and allowing events with three or more leptons instead of exactly four. Crucially for the
low-mass region, this search uses a full particle-flow form of isolation, and does not count
leptons towards each others’ isolation cones. We include a Z-veto to help reduce Z+jets
and diboson backgrounds. We also focus on “below-Z” events, where the `+`− pair closest
to the Z mass is below 75 GeV. These vetoes have little effect on the signal efficiencies.27
In reconstructing the µ+µ− resonance, there remains a combinatoric issue when more
than one pairing of this type is possible. This ambiguity afflicts the majority of 3-lepton
and 4-lepton events containing at least one µ+µ− pair, since muons are reconstructed with
higher efficiency than electrons. (E.g., µ+µ−µ± is found more often than µ+µ−e±.) In
practice, it is possible to pick the smallest-mass pairing for ma � mh/2 and the largest-
mass pairing for ma ' mh/2. However, for ma ' mh/4, neither of these options is ideal.
Instead, we can construct a third option by using the fact that mh ' 125 GeV, that the
Higgs decays isotropically, and that it is usually produced with little transverse boost: we
27 It might also be possible to apply a E/T discriminator for this search, though we have not attempted this.
The E/T in signal events tends to be below 50 GeV. An accurate understanding of the efficacy of a E/T
cut would require a resolution model, as well as a model for the E/T distribution of backgrounds. An
approximate reconstruction of the Higgs resonance might also be possible, and usable either for further
discrimination or for verification of the source of a possible signal.
90
pick the µ+µ− pair whose trajectory would make the largest opening angle with the beam
in the Higgs rest frame, assuming pT (h) = 0. For each ma, we use the pairing choice that
gives the strongest resonance peak.28
Estimating backgrounds to such a search can be difficult, as leptons from heavy flavor
decays and from fakes can be significant contributions. We have simulated the contributions
from electroweak 3-lepton and 4-lepton production, including taus and allowing for Z∗/γ∗
down to m ∼ GeV. Given a signal that lives inside of a resolution-limited mass window of
approximately (1± 0.01)ma, these backgrounds are usually small, tallying to O(1 event) for
any ma for 2012. The dominant Z∗/γ∗+jets background can be coarsely estimated from the
sum of “below-Z” bins of analysis (1), and would constitute approximately 800 events for
m(µ+µ−) ∼> 10 GeV with 20 fb−1. (In this estimate, we conservatively do not attempt to
remove the e+e− events.) We are not given a spectral shape for this background, but if we
assume that it is not very strongly-featured, then we can estimate O(10 events) per 1 GeV
interval. We also do not know the spectrum for m(µ+µ−) ∼< 10 GeV, though the shrinking
absolute resolution on m(µ+µ−) (down to less than 100 MeV at CMS) allows the differential
background rate to grow by an order of magnitude without affecting S/B. Of course, extra
care would need to be taken in the vicinity of known hadronic resonances such as the Υ’s.
To give a sense of what might be possible with the 2012 data set, we show in Fig. 25
the limits assuming a sequence of possible background levels with m(µ+µ−) within ±1% of
ma, and neglecting systematics. Taking as reference Br(h → 2a → 4τ) = 10%, the signal
rates inside the peak vary from 8 events for ma = 4 GeV, to 25 events for ma = 60 GeV.
Depending on the background assumption and on ma, the excluded Br(h→ 2a→ 4τ) varies
from percent-scale to just above 10%. This strong level of exclusion applies even down to
ma ' 2mτ .29 We imagine that these results will only improve as data from the next run of
the LHC becomes available, provided that the multilepton triggers can be maintained at pT
thresholds comparable to their 2012 values.
28 The crossover between smallest-mass and largest-mass choices being the most effective is at ma ' 40 GeV,
and in this region the largest-opening-angle choice keeps about 15% more events in the peak. For very
low-mass resonances, this choice underperforms the smallest-mass choice by a comparable amount, and
similarly for high-mass resonances (near mh/2) relative to the largest-mass choice.29 Note that while isolation of a single lepton from the a → τ+τ− side of the event becomes progressively
more difficult for low-mass points, Br(a → µ+µ−) is also increasing. At 4 GeV, the rate has doubled.
This effect would be even more pronounced for CP-even scalars.
91
5 15 25 35 45 55ma in GeV
2
44
6
8
10
12Br(h→aa→
4τ)(%
)
Expected exclusion with 2012 data from µ+ µ− resonancesB=0B=5B=10B=20
FIG. 25: Median estimates of expected indirect exclusions on Br(h → 2a → 4τ) using the sub-
dominant (a → 2τ)(a → 2µ) channel and exploiting that leptonic branching fractions of a are
mass-ordered. The results are based on a simulated µ+µ− resonance search in ≥ 3` events, assum-
ing the 2012 data set. Since we cannot reliably predict the background under the resonance peak,
we show expected exclusions for B = 0, 5, 10 and 20 events respectively. We neglect systematic
uncertainties. (The lowest displayed mass is 4.0 GeV.)
7. h→ 4j
Standard Model decays of the Higgs boson can lead to a four-jet final state via interme-
diate vector boson decays, h → WW ∗/ZZ∗ → jjjj. Only one of the jet pairs is produced
on-resonance in this process. In this section, we discuss the distinct possibility of exotic
Higgs decays to 4j in a two-step decay process proceeding through a neutral (pseudo-)scalar
field a: h → aa → jjjj. There are then two jet-pair resonances. Below, we outline the
theoretical motivations for considering 4j decays of the Higgs, and discuss the LHC phe-
nomenology and future discovery prospects of this channel.
92
7.1. Theoretical Motivation
The h → jjjj channel has been extensively studied in the context of super Little Higgs
models [306–308] (a brief description of the Little Higgs mechanism is given in §1.3.9). The
intermediate decay product, a, is a PNGB and generally very light. In a large region of
parameter space of these models, h→ aa→ jjjj is the dominant decay mode.
Given that the Higgs mass of approximately 125 GeV requires fine-tuning of the simplest
versions of these models, one may take a simplified model approach for the cascade decay
in the presence of a light pseudoscalar (or scalar), a. Two possibilities allow for the decay
of a to jets:
(i) The pseudo(scalar) a can mix with another heavier pseudoscalar if a second Higgs
doublet is present, for example in the NMSSM or, more generally, in the 2HDM + S models,
see §1.3.2 and §1.3.6). This allows for the decay of a to SM fermions, often (depending on
the 2HDM Type) dominated by a→ bb for ma > 2mb and a→ τ−τ+ for 2mτ < ma < 2mb
for a large or moderate tan β. This leads to 4b, 2b2τ , 2b2µ, 4τ , and 2τ2µ signals as discussed
in §3, §4, §5, and §6. However, if a is very light (3mπ < ma < 2mτ ), it predominantly decays
to two (merged) light jets as the above channels are not kinematically viable.
If tan β is small (tan β . 0.5), the couplings of a to the down type quarks and charged
leptons can be very suppressed. In this case, a dominantly decays to light (mostly charm)
jets even if decays to b’s or τ ’s are kinematically allowed. Thus, the parameter space of ma
up to mh/2 is available for the exotic decay mode. A similar situation also occurs in the
“charming Higgs” scenario of the Little Higgs model [219].
(ii) New heavy BSM vector-like fermions can couple to a and, therefore, allow for its
decay into gluons or photons through loop processes [201, 262, 309]. This scenario can be
realized in Little Higgs models and extra dimensional models. For ma above a few GeV
up to mh/2, h → aa → gggg dominates over h → aa → γγgg and h → aa → γγγγ. In
general, the signal is hard to find against combinatorial background. However, large masses
of the new vector-like fermions may lead to visibly displaced vertices of a→ gg, which can
enhance the discovery potential of the channel [309]. Studies on related decay modes in this
scenario, h→ aa→ γγgg and h→ aa→ γγγγ, can be found in §8 and §9, respectively.
93
7.2. Existing Collider Studies
Before the discovery of the 125 GeV Higgs boson, much of the phenomenology of the Higgs
decaying to four jets was aimed at hiding the Higgs boson at LEP. One way to accomplish
this was in the “buried Higgs” scenario , where the decay h→ jjjj is “buried” in the large
QCD background. Indeed, the LEP bounds for this scenario are much weaker than the
bound on a SM Higgs. For mh > 90 GeV [289], ma was studied in a range where each pair
of jets from the pseudoscalar decay would be highly collimated and appear as a single jet.
There are a few existing collider studies for the 14 TeV LHC run in the four-jet final
state. In [220] the authors study the h → 4c decay mode in the context of “charming
Higgs”. We mention this study here since it does not use b-tagging and hence useful for
generic 4j decays. The study uses jet substructure to help identify the pseudoscalar as a
boosted jet while reducing the otherwise overwhelming background.
Other relevant collider studies are [267] and [218], which we briefly summarize below.
(There also exist collider studies that consider exotic Higgs production modes [266], but we
do not consider them here.)
In [267], Higgs production in association with a W boson is considered as the production
mode for mh = 120 GeV followed by the Higgs decay, h → aa → jjjj. The pre-selection
cuts in this analysis include isolated leptons with pT > 20 GeV, at least two jets with
pT > 40, 30 GeV, reconstructed leptonic W transverse mass mT < mW , and a b-jet veto to
reduce SM background. Further analysis is divided into categories depending on the mass
of a:
• ma = 4 GeV : In this case the gluons from a decay appear as a single jet to the
HCAL. ECAL variables are imposed to distinguish these merged jets from single-
pronged QCD jets. 7σ significance is possible at the LHC14 with 30 fb−1 data assuming
Br(h → aa → gggg) ∼ 100%. However, assuming a more realistic branching ratio of
Br(h→ aa→ gggg) ∼ 10% in the post Higgs discovery era, 2σ exclusion (3σ evidence)
is possible with 300 fb−1 (500 fb−1) of data at LHC14.
• ma = 8 GeV : Simple jet substructure techniques can be used for discovery. The
authors find that ∼ 3σ statistical significance can be reached with 30 fb−1 data as-
suming Br(h→ aa→ gggg) ∼ 100%. With Br(h→ aa→ gggg) ∼ 10%, however, 2σ
94
exclusion (3σ evidence) requires 1000 fb−1 (3000 fb−1) of data at LHC14.
A separate jet substructure analysis on h → aa → jjjj is also presented in [218], with
the inclusion of the tth production channel besides the V h channel, demonstrating similar
discovery potential in both channels. Here variables sensitive to the soft radiation patterns of
the color singlet a→ gg jet are employed instead of ECAL-based observables. The authors
reach a similar conclusion for discovery prospects as described above.
The above two analyses [218, 267] have exploited the fact that very light (pseudo)scalars
are boosted, leading to two fat-jets. A more recent study [265] explores the ma > 15 GeV
regime. It focuses on the substructure of fat-jets containing an entire boosted Higgs decay,
and that could be 2-, 3-, or 4-pronged. As before, Higgs production in association with
vector bosons is considered. The authors include two cases depending on the mass of the
scalar, s: (i) light scalar (15 < ms < 30 GeV) and (ii) heavy scalar (30 GeV < ms < mh/2).
In the lighter regime, the h → ss → jjjj signature with 100% branching ratio can be
observed at a significance of 3σ with 100 fb−1 of 14 TeV LHC luminosity, while for the heavy
scalar case, the significance is too small to observe with the same amount data. For a more
realistic Br(h → 2a → 4j) = 10%, 2σ exclusion for the light scalar case requires 1500 fb−1.
(Note the achievable limits become much stronger for h→ 4b with b-tags, see §3.)
7.3. Existing Experimental Searches and Limits
There are currently no existing experimental searches looking for a four-jet resonance in
the low invariant mass region, which is understandable due to the large QCD background.
Neither are there any existing searches that look for fat-jet resonances.
Overall, this is a highly challenging exotic Higgs decay channel. For ma . 5 GeV,
2σ exclusion of Br(h → 2a → 4j) = 10% requires 300 fb−1 of LHC14 data, while ma &
5 GeV requires more than 1000 fb−1. This search should be undertaken at the 14 TeV LHC
(especially for light ma, where the decay is particularly motivated), but it is not plausibly
part of the LHC7 or 8 physics program.
95
8. h→ 2γγγ2j
A relatively clean exotic decay mode of the Higgs boson is h→ 2γ2j [310]. The SM rate
for this signature is negligible: decays into 2γ2q are highly Yukawa suppressed while the
2γ2g process is loop induced. However, going beyond the SM, more possibilities arise. In
particular, here we consider Higgs boson decays to two scalars ss(′) which subsequently decay
into photons and gluons or quarks. Below we outline some possible theoretical scenarios
leading to such decays and briefly discuss their collider phenomenology.
8.1. Theoretical Motivation
There are several ways in which a SM singlet scalar decays to photons, gluons or quarks.
For example, it can do so via mixing with the Higgs boson, as in the singlet extensions
discussed in §1.3.1 and §1.3.2. This will generally give a very suppressed rate to photons
compared with that of quarks or gluons, due to the electromagnetic loop factor.
Alternatively, a singlet scalar s may couple to gluons and photons via a dimension-5
operator sF µνFµν , which arises by introducing new colored and charged vectorlike states
and coupling them to s. Such scenario can easily accommodate larger or even dominant
s → 2γ branching ratios, depending on the color vs. electric charge assignments of the
new states. As a simple example, consider adding new heavy Dirac fermions ψi along with
Yukawa couplings of the form λisψiψi. The fermions reside in a representation Ri under
SU(3)C , have electric charge Qi and mass mi. The scalar s then decays to gluons and
photons via heavy fermion loops. The resulting branching ratios satisfy
ρ =Br(s→ 2γ)
Br(s→ 2g)=
1
8
(α
αs
)2 [∑λi Q
2iN(Ri)/mi∑
λi C(Ri)/mi
]2
, (75)
where N(Ri) and C(Ri) are the dimension and normalization factor of the representation
Ri (the normalization factors of the lowest lying color representations R = 3, 6, 8 are C =
1/2, 5/2, 3). For example, one heavy down-type quark b′ and one heavy charged lepton τ ′ (a
combination which appears in a single ‘5’ multiplet of SU(5), along with a heavy neutrino),
with masses m2 and m3, and Yukawa couplings λ2 and λ3, respectively, would result in
ρ =1
18
(α
αs
)2(1 + 3
λ2
λ3
m3
m2
)2
' 0.02
(λ2
λ3
)2 ( m3
30 TeV
)2(
10 TeV
m2
)2
. (76)
96
Note that the heavy fermions need not be light in order to induce 2γ or 2g decays, as long
as the singlet s does not mix with the Higgs boson.
In principle, the 4γ mode (§9) is much cleaner than 2γ2j, which is in turn much cleaner
than the very difficult 4j (§7). However, since
Br(h→ 4γ)
Br(h→ 2γ2g)' 1
4
Br(h→ 2γ2g)
Br(h→ 4g)' 1
2
Br(s→ 2γ)
Br(s→ 2g)=ρ
2, (77)
for small enough values of ρ, as defined in Eq. (75), the 4γ rate would be too small to be
observable for a given integrated luminosity. In such a situation, which occurs if b′ and τ ′
are degenerate in mass and couplings, the 2γ2j signature may be competitive with 4γ.
Of course, the model described above is just one example of h → 2γ2g decays. Other
examples may feature two different states, s and s′, allowing for even more model-building
freedom, or decays to quarks instead gluons. Since the main focus of this section is to explore
the 2γ2j signature and propose ways to discover it at the LHC, we content ourselves with
the model described above and continue to discuss discovery reach and limits.
8.2. Existing Collider Studies
In [310], a search has been proposed for this channel, and the discovery (5σ) reach at
the 14 TeV LHC with 300 fb−1 was derived as function of the scalar mass ms and Higgs
mass mh. Gluon fusion (ggF) and W -associated production (Wh) were considered. Here
we only make use of the latter, both because it provides superior sensitivity in this analysis
and because the ggF study, which was conducted before the LHC came online, incorporated
di-photon pT thresholds which are much lower than current triggers.
The Wh analysis in [310] proceeds as follows: events are required to contain one lepton,
two photons and two jets with pT > 20 GeV and |η| < 2.5 for each of these objects.
Moreover, each object pair (jj, γγ, jγ, j`, `γ) is subject to an angular isolation criterion of
∆R > 0.4. The events are also required to have E/T > 20 GeV. Additional cuts made
were ∆φγγ < 1.5, ∆φjj < 1.3, and |mjj −mγγ| ≤ 15 GeV. The Higgs mass resolution was
assumed to be ∼ 8 − 10 GeV. The signal efficiency is claimed to be between 3% and 15%
within the relevant mass range.
Rescaling the 5σ limit at 14 TeV with 300 fb−1 to 95% CL yields the sensitivity shown as
97
8 TeV est.
14 TeV H100 fb-1L
14 TeV H300 fb-1L
pp ® Wh , 95% CL Upper Limit
20 30 40 50 60
0.010
0.100
0.050
0.020
0.200
0.030
0.015
0.150
0.070
ms @GeVD
BrH
h®
2Γ2gL
FIG. 26: Projected 95% CL limits on the branching fraction for h→ 2γ2g in associated production
(Wh), as function of ms. The blue curves refer to 300 fb−1 (solid) and 100 fb−1 (dashed), both at
the 14 TeV LHC. The dashed-dotted green curve shows a conservative estimate of the sensitivity
for 20 fb−1 at 8 TeV. All three limits build on the proposed search in [310] (300 fb−1 at 14 TeV
LHC), by scaling background with luminosity but not changing its cross section, while signal is
rescaled according to both luminosity and cross section. This underestimates the achievable 8 TeV
limit. See text for more details.
the solid blue curve in Fig. 26. An estimate for the lower luminosity30 of 100 fb−1 is shown
as the blue dashed curve. At the 14 TeV LHC, a sensitivity to Br(h→ 2γ2j) below 0.01 is
possible for part of the kinematically allowed s mass range. This study can also be used to
obtain a conservative estimate of the sensitivity at the 8 TeV LHC. We scale the production
cross section down appropriately without doing so for the background cross section. This
will underestimate the strength of the limit (assuming the efficiencies do not change by a
large amount at 8 vs. 14 TeV). The resulting 95% CL sensitivity is shown as the green dash-
dotted curve in Fig. 26. Run I data should be able to set a limit on Br(h → 2a → 2γ2j)
as low as ∼ 0.04 for some scalar masses, and likely better than that, given our pessimistic
rescaling.
Two comments are in order:
1. Note that the angular isolation cuts reduce the background, but effectively eliminate
sensitivity for ms . 20 GeV. This weakness of the proposed search might be remedied
30 Our rescalings include the assumed 10% systematic errors on the background rate [310].
98
by means of jet substructure-inspired techniques [311, 312] (see also §9).
2. Since the best limits seem to be given by associated Wh production, we do not expect
too much difficulty with triggering. However, since the threshold for the single lepton
trigger will be raised for LHC14, it would be helpful to have a trigger that requires a
lepton and a photon.
8.3. Existing Experimental Searches and Limits
There are no limits from existing searches. Potentially relevant searches, such as super-
symmetry searches and isolated photon-pair searches [313, 314] are generally insensitive to
h → 2γ2j, since (a) they employ relatively hard cuts and (b) without a cut on the total
invariant mass, the QCD background is overwhelming.
The h→ 2γ search in the VBF mode also cannot be used to place limits on 2γ2j, since
the VBF dijet tag is targeted at the forward and high rapidity gap region where a 2γ2j
signal is faint.
8.4. Proposals for Future Searches
Based on the results from [310], both the gluon fusion and the V h associated production
mode should be explored for h→ 2j2γ sensitivity at LHC Run I and II.
An interesting issue arises for very light intermediate resonances, which may result in
unexpected signatures, as follows. As mentioned above, the previous search strategy involved
an isolation cut on the photons. This spoils sensitivity for light s particles, since these
would decay to a collimated pair of photons or gluons. One would therefore be missing an
important portion of parameter space below ms ∼ 20 GeV. Using more sophisticated photon
identification inspired by jet substructure techniques will improve the situation. However,
for low enough ms . GeV , the two jets cannot be resolved, resulting in a j+ 2γ signature.
Furthermore, in [315] it was shown that for very low ms . 100 MeV the diphoton system
is so collimated that a substantial fraction of the photon pairs would deposit their energy
in a single electromagnetic calorimeter cell,31 resulting in h → 4γ mimicking 2γ and 3γ
31 The study in [315] was geared toward the ATLAS detector, but similar principles may be applied to CMS
99
signatures. While a scalar as light as to induce merged photons is generally not able to
decay into gluons (namely, hadrons), having two different states with different masses may
allow for merging both photons and gluons, resulting in signatures such as 2j + γ or j + γ.
It is therefore interesting to consider such topologies, although they are considered “im-
possible” for Higgs decays due to the “wrong” quantum numbers they seemingly possess.
These subtleties should be taken into account when conducting a future 2γ2j search. At
the trigger level the two merged photons could pass as one single photon, necessitating the
use of a single photon (possibly + jets) trigger.
9. h→ 4γγγ
Here, we consider the decay of a Higgs to four photons. In the SM, the branching fraction
for this decay is negligible, as it results from a dimension-nine operator and contains an
additional factor of α in the amplitude relative to h→ γγ. However, it can be important in
certain new physics scenarios, as we now discuss.
9.1. Theoretical Motivation
The basic decay chain that we consider is h → aa(′), a(′) → γγ. Enumerating the
possible quantum numbers of the intermediate particles is simple if they decay into two
photons and have spin less than two: they must be neutral and spin-0 by the Landau-Yang
theorem [316, 317]. The CP phase of the a(′) makes no difference phenomenologically as
long as the photon polarizations are not measured.
There are a number of theoretically well-motivated candidates for a, among them the
lighter pseudoscalar of the NMSSM, any pseudoscalar that mixes with the CP-odd Higgses
of the (N)MSSM, or a generic SM-singlet boson whose coupling to photons is mediated by
a renormalizable coupling to heavy vector-like matter. In the first two cases, the coupling
of a to light SM fermions can make the branching for a → γγ subdominant, but the low
backgrounds in 4γ can nonetheless make it an interesting final state. On the other hand, if
a couples at the renormalizable level only to the Higgs and to heavy vector-like uncolored
matter, it may only be able to decay to γγ, rendering the 4γ final state extremely important.
as well.
100
If, alternatively, the vector-like matter is colored and a→ gg is allowed, h→ ggγγ can also
be important (see §8 for details).
It is also worth noting that if ma < 2mµ, only the γγ and e+e− final states may be
kinematically allowed. The other final states in this case, 4e or 2e2γ, are broadly similar
phenomenologically to 4γ, since they involve electromagnetically interacting particles. We do
not discuss them further here, leaving a detailed study for the future [318]. Furthermore, as
we show below, for ma . 100 MeV with a decaying only to photons, a is typically long-lived
on collider scales, potentially leading to displaced vertices or missing energy. Long-lifetimes
are also possible in certain hidden valley models, even for much larger masses [31, 222].
9.2. Existing Collider Studies
The h → aa → 4γ decay chain was studied in [201], focusing on the Tevatron. In
this paper, it was pointed out that for ma . 0.025mh the a’s are boosted enough that
photons coming from their decays are collimated to the extent that they will often deposit
their energy in a single calorimeter cell, fail isolation cuts and potentially be reconstructed
as a single photon. (We discuss some of the experimental issues regarding closely-spaced
photons below, focusing on the LHC.) This light a scenario is motivated if, e.g., a is the
lightest pseudoscalar in the R-symmetric limit of the NMSSM (see §1.3.7). The results of
the analysis of [201] imply that the full Tevatron dataset is sensitive to branchings of h→ aa
at about the 0.5% level or larger, assuming Br (a→ γγ) = 1.
In [309], a detailed study was performed of the h → aa → 4γ decay at the LHC with√s = 14 TeV. The experimental cuts made in this study were that the transverse momenta of
the photons were all greater than 20 GeV, the distance between the photons was ∆R > 0.4,
the photons had rapidity |η| < 2.5, and there were two separate pairs of photons that
reconstructed the same invariant mass (the candidate a mass) to within 5 GeV. Finding
backgrounds to be negligible with these cuts, this work indicated that for a Higgs at 125 GeV,
300 fb−1 of data at the 14 TeV LHC would allow branchings Br (h→ aa) ' 5× 10−5 to be
discovered at the 5σ level for 10 GeV . ma . mh/2, assuming that the a’s decay promptly
to photons only. Rescaling this to 100 fb−1 would indicate that Br (h→ aa) ' 9 × 10−5
could be found at 5σ. The isolation cut of ∆R > 0.4 is the reason for the lower bound
on the a mass that can be accessed. A naive rescaling by the decreased luminosity and
101
Higgs production cross section of the 7 and 8 TeV datasets, assuming that the dominant
backgrounds’ cross sections do not change appreciably, implies that the current data is
sensitive to Br (h→ aa) ∼ few × 10−4. As emphasized in [309], the reach is extremely
sensitive to the value of the photon pT cut, especially in the case of a relatively light Higgs
with mh = 125 GeV.
Closely-spaced pairs of photons in h→ aa→ 4γ at the LHC when ma � mh = 125 GeV
were studied recently in [315], motivated by early hints at√s = 7, 8 TeV that the Higgs
rate to diphotons could be larger than in the SM. However, photon pairs that fail mutual
isolation criteria might or might not be detected as a single photon depending on the details
of their geometric distribution, as we now explain in detail.
As mentioned above, it was noted in [201] that at the Tevatron, for sufficiently small ma,
the pairs of photons from each a decay could be collimated enough to appear as a single
photon in the detector. If ma ∼< 10 GeV with the a’s produced in the decay of a 125 GeV
Higgs, the photons that they decay into will fail the typical isolation cut of ∆R = 0.4.
However, their energy depositions in the ECAL will normally be broader than that of a true
single photon (whose electromagnetic shower has a typical width that is material-dependent,
called its Moliere radius) and will not be tagged as a single photon. As the mass of the a
is pushed down further, the decay photons do eventually become merged enough that their
energy depositions are no longer much broader than a single photon’s. The value of ma
where this becomes important depends on the spatial resolution of the ECAL in question.
The increased granularity of the LHC detectors compared to those at the Tevatron means
that ma must be smaller at the LHC than at the Tevatron for this to be the case. At
ATLAS, a single photon’s electromagnetic shower deposits its energy in several neighboring
cells in the innermost central portion of the ECAL where the cells have a width in the η
direction of 0.0031 (corresponding to ∼ 0.5 cm) because the Moliere radius of the absorbing
material, lead, is O(cm) [319]. In Ref. [315], it was found that requiring ∆η < 0.0015
(half the smallest cell size at ATLAS) between the two nearby photons from an a decay
successfully reproduced the shower shape cuts used to distinguish single photons. For the
photons to be this closely separated, ma . 100 MeV.32 In such a case, an apparent increase
32 This critical value of ma makes sense since the LHC detectors were designed to be able to tell neutral
pions apart from single photons.
102
of ∼50% in the apparent h → 2γ rate could be achieved for Br (h→ aa) ' 10−3 to 10−2.
Other possible experimental consequences of this scenario mentioned in [315] are an increase
in the number of events containing a converted photon, a mismatch between the momentum
of charged tracks and the energy deposition in the calorimeter in conversions (when one of
the two nearby photons converts), or the appearance of apparent h →“γ + j” events when
one pair of photons is very collimated, faking a single photon, while the other is broader,
failing isolation requirements for photons and looking like a jet (with large electromagnetic
content).
Additionally, the usefulness of jet-substructure-motivated detector variables in distin-
guishing closely-separated photons (termed photon-jets generically in [320]) from single
photons and their interplay in h → 4γ faking h → γγ at the LHC was studied in de-
tail in [311, 312], dealing with both the case where the photons were merged enough to
potentially fake a single photon and that in which they are less closely merged but do still
fail isolation cuts, potentially looking like a jet. Examining h → 4γ with a Higgs mass of
120 GeV, they determined that the use of such variables could decrease the rate of photon-
jets faking single photons by a factor of over 10 while preserving at least 80% of the single
photon signal.
Most of the literature assumes that the photon pairs necessarily reconstruct two equal-
mass resonances, however this will not be the case when two different particles a and a′ are
introduced and the decay mode h→ aa′ is allowed. For an example of such a model which
assumes ma ≈ mh and ma′ ∼< GeV, which was originally designed to increase an observed
h→ γγ rate, see [321]. In general, there are no direct constraints on ma, ma′ .
We pause here to note that if a or a′ is light, it is quite natural to get a decay length that
is detector-scale. For example, parametrizing the coupling of a pseudoscalar to photons as
L =πα
MaFµνF
µν (78)
one gets a decay length, if they are produced in the decay of h at rest, of
γcτ ' 0.75 cm
(M
5 TeV
)2(1 GeV
ma
)4 ( mh
125 GeV
). (79)
It is easy to see that for ma . 100 MeV and M & 1 TeV,33 a’s decay length could be
33 We would expect such a scale if a’s coupling to photons came from integrating out charged matter above
the electroweak scale.
103
of the order of several meters. Long decay lengths are therefore a generic feature of light
pseudoscalars decaying to photons and should be kept in mind when contemplating such
signals.34
9.3. Existing Experimental Searches and Limits
A search for h → aa → 4γ in the case where ma � mh leading to very collimated pairs
of photons was performed by ATLAS on 4.9 fb−1 of 7 TeV data [322]. The search was very
similar to the standard one for h→ γγ but shower shape variable cuts were relaxed to allow
for increased acceptance of the 4γ signal. This resulted in a very good acceptance for events
coming from the h → γγ channel. Results were presented for ma = 100, 200, 400 MeV,
limiting Br (h→ aa) Br (a→ γγ)2 . 0.01 at mh = 125 GeV.35 For larger a masses, there
are no limits from collider searches.
Results from low energy experiments (see, e.g. Ref. [132]) are not constraining on this
scenario for ma & 10 MeV so long as the a’s decay promptly at the LHC [315].
9.4. Proposals for New Searches at the LHC
A search for h → 4γ using the full 7 and 8 TeV dataset of both experiments would
be highly desirable. Reference [309] indicates that 300 fb−1 of data at the 14 TeV LHC
can access values of Br (h→ aa) Br (a→ γγ)2 > 5 × 10−5 at 5σ for ma & 10 GeV. For
ma . 10 GeV, the 4γ signal can be hard to disentangle from the large QCD dijet background
and for ma . few × 100 MeV it can even look very similar to h → γγ. In these cases, as
shown in [311, 312], using detector variables from jet substructure can greatly reduce the
QCD dijet backgrounds and help to distinguish these final states, greatly increasing the
reach for h → 4γ. Thus far, most work on this signal has concentrated on either the very
light a regime where two photon pairs are very collimated or where ma > 10 GeV and the
four photons are well separated. The intermediate mass region is also well motivated and
we encourage it to be studied as well.
34 This conclusion can be modified slightly when other decay channels for a are present or if the operator
aFµν Fµν is generated below the electroweak scale. See [315] for details.
35 In the SM Br (h→ γγ) ∼ 2 × 10−3. Therefore the impact of the SM diphoton channel on this bound is
still rather small.
104
The assumption that the two intermediate particles have the same mass cuts down on
backgrounds but a more general search strategy looking for γγ bumps in h→ 4γ could help
to shed light on a scenario where this decay is dominantly mediated by two particles with
distinct masses.
Lastly, macroscopic decay lengths for the particles mediating h → aa(′) → 4γ can be
naturally realized in simple models, especially when they are light or if they are composites
from a hidden valley, which motivates searches for 4γ events where two pairs of photons
each resolve displaced vertices.
10. h→ ZZD,Za→ 4`
Below we discuss decays of the form h→ Z+X, where X denotes a non-SM light boson.
We focus on two possibilities:
1. X = ZD, a new gauge boson that acquires a mass and mixes with the SM gauge
bosons, see §1.3.5.
2. X = a, a light pseudoscalar as in the 2HDM+S and the NMSSM [189], see §1.3.2,
§1.3.7.
In both cases we are interested in a two-body decay of the Higgs boson, meaning we require
MX . 34 GeV. We outline the theoretical motivation to consider such decays and discuss
the limits by LEP, Tevatron, and LHC.
10.1. Theoretical Motivation
10.1.1. h→ ZZD
As discussed in §1.3.5, many theories feature a hidden U(1) sector with small kinetic or
mass mixing the the SM photon and Z-boson. This possibility often arises in connection
to dark matter, but similar phenomenology can also arise in more general hidden valley
models, see §1.3.10. The minimal setup Eq. (38) to generate h → ZZD decay involves a
kinetic mixing term between the hypercharge gauge boson and the dark U(1) gauge boson
Lgauge ⊃1
2
ε
cos θWBµνZ
µνD , (80)
105
where hatted quantities are fields before their kinetic terms are canonically renormalized by
a shift of Bµ. In the canonical basis, SM matter has a dark milli-charge and there is mass
mixing between the SM Z-boson and ZD. The dominantly dark vector mass eigenstate has
photon-like couplings to SM fermions (proportional to the small mixing ε) up toO(m2ZD/m2
Z)
corrections, see Eq. (47). If ZD is the lightest state in the dark sector it will decay to SM
fermions via this coupling. Prompt decay requires ε & 10−5 − 10−3 (depending on mZD),
and the largest Br(h→ ZZD) allowed from indirect constraints is ∼ 10−3, see Fig. 12.
It is also possible to have pure mass mixing after EWSB via operators of the form hZµZ ′µ,
but in this case additional constraints from parity violating interactions and rare meson
decays apply, see [164, 165, 170]. Generically, new physics similar to that which generates
kinetic mixing may also generate dimension-6 terms of the form H†HBµνZDµν/Λ2. Once
the Higgs acquires a VEV, this term yields the coupling in Eq. (80).
10.1.2. h→ Za
Next we consider the decay h→ Za. This is motivated by, for example, the 2HDM+S or
the NMSSM, where one of the CP-odd Higgs masses can be small. The relevant interaction
Lagrangian in terms of mass eigenstates h and a is given by Eq. (18) with an additional
Yukawa term:
Lint = g(a∂µh− h∂µa)Zµ − gaf iγ5fa. (81)
with g =√
(g2 + g′2)/2 sin(α − β) sin θa. The parameter α is the mixing angle between
the doublet scalars, tan β = vu/vd, and θa is the mixing angle between the uneaten doublet
pseudoscalar A and the singlet pseudoscalar. Since the Higgs coupling to ZZ and W+W− is
also proportional to sin(α− β), the SM-like rates in those channels (as well as the diphoton
mode) favor the decoupling limit α = π/2 − β. θa can be constrained by direct LEP
and Tevatron searches for the CP-odd Higgs, but the SM-like Higgs could still have large
branching fractions to Za [189]. The pseudoscalar coupling to fermions can be extracted
from Table II,
ga = sin θa tan βmf
v, for b, τ, and µ (82)
and the overall size of θa does not affect its branching ratios.
For the length of the LHC program it will likely be safe to take Br(h→ Za) = 10% as a
benchmark point. In the next section, we discuss the experimental constraints on this mode.
106
Depending on the mass of this pseudoscalar, the dominant decay mode could be bb, τ+τ−,
or µ+µ− (ss). We consider all of these cases when proposing search strategies.
10.2. Existing Collider Studies
Up to different branching ratios and some angular correlations the final states for h →
ZZD and h→ Za are identical. As such, collider studies and experimental searches for one
channel generally apply to both. The two relevant parameters to define a simplified model
for this channel are
mX and Br(h→ ZX → Zyy) (83)
for X = a, ZD and y = some SM particle, where the different a, ZD branching ratios lend
different importance to different choices of y.
There have not been many collider studies specifically performed for the h→ Za mode.
Ref. [189] pointed out that this channel may be very large in the context of the NMSSM.
Ref [323–325] discussed heavy non-SM-like Higgs decaying into Za.
More searches have been inspired by looking for a ZD. The phenomenology of a ZD
with mass mixing to the Z has recently been discussed in [164, 165, 170, 326] (see also,
e.g., [30, 166, 168, 327] for earlier work), including collider phenomenology of h → ZZD,
h→ γZD, and h→ ZDZD decays, as well as low energy constraints from colliders and fixed-
target experiments, g − 2 of the muon and electron, rare meson decays, and electroweak
precision observables (see §11 for the h→ ZDZD mode).
In [165], the authors designed a search for pp → h → ZZD → e+e−µ+µ−. The back-
grounds considered are Z(→ `+`−)jj, j faking ` (probability ∼ 0.1%) and leptonic tt (re-
ducible), as well as h → ZZ∗, Zγ∗, ZZ → 4` (irreducible). The authors of [165] assumed
only mass mixing of the form εZm2ZZ
µZDµ. For mZD ∼ 5 − 10 GeV, they find that the
14 TeV LHC has 2σ sensitivity to Br(h → ZZD → Z``) ∼ O(1) × 10−4 with 30 fb−1 of
luminosity.
10.3. Existing Experimental Searches and Limits
A light pseudoscalar a can be searched for in Υ decays at Babar [328], top decay at the
Tevatron [329], and direct single production and decay to dimuons at the LHC [330, 331].
107
These dedicated searches are discussed in other sections of this document, and their reach
depends on many parameters of the theory. There are also many constraints (most of them
not from high energy colliders) on the existence of a ZD, see Fig. 12, but there are large
regions of parameter space relevant for exotic Higgs decays that are not excluded.
Our focus is the hZX vertex (X = a, ZD). No direct search for h→ Za or ZZD has been
performed to be best of our knowledge, but there are several channels and other searches at
LEP, Tevatron, and LHC that are sensitive to this interaction term.
LEP
The hZX vertex not only gives rise to the h → ZX decay, but also opens the channel
e+e− → Z∗ → hX at LEP. Related searches include e+e− → ha, ZZ ′ → 4b [332], 4τ [332]
and 2b2τ [332]. For Br(h → Za) = 10%, these searches are not constraining because
the cross section for e+e− → Z∗ → ha is at the sub-fb level. Even without considering
any branching fraction suppression to the final states, LEP’s integrated luminosity is still
too small to be sensitive. One can also imagine more spectacular production modes such
as e+e− → ha → aaa → 6b and e+e− → ha → aaa → 6τ , which can be recast into
e+e− → ha→ Zaa→ 6b and e+e− → ha→ Zaa→ 6τ . These channels yield no constraints
even before taking into account kinematic acceptances.
Tevatron and LHC
The most relevant existing search sensitive to h → ZZD and h → Za is h → ZZ∗ → 4`
by CMS [176] and ATLAS [177], where 4` stands for electrons and muons. The clean 4`
decay makes these existing searches very sensitive to ZZD or Za decaying into leptons.
The leptonic h→ ZZ∗ searches divide the four leptons of each event into two pairs, the
“leading” pair (likely to have come from an on-shell Z) and the “subleading” pair (from the
off-shell Z∗, denoted sometimes as “Z2” or m34). The subleading dilepton mass distributions
from ATLAS and CMS are shown in Fig. 23 of [177] and Fig. 9 of [176], respectively, using
the full 20 + 5 fb−1 data set of LHC7+8. With this information it is easy to estimate limits
on h→ ZX decay.36 The new state X will contribute to h→ Z`` events in two ways, firstly
through resonant h → ZX production, and secondarily through interference with the SM
amplitude h → ZZ∗. Here we consider only resonant production, obtaining a conservative
36 The `+`− distribution in h → Z`` events can also be used to search for indirect effects of new physics
above the Higgs mass [333, 334].
108
estimate on Br(h → ZX); a study incorporating the off-shell contributions will appear in
future work.
A ZD or a decaying through some small mixing to SM particles will have a much smaller
width than ΓZ ≈ 2.6 GeV or ΓhSM≈ 4.07 MeV. Given the . 3% dilepton mass resolution
of the experiments and the subleading dilepton mass (MZ2) binning of 1.25 (2.5) GeV by
CMS (ATLAS) it is safe to assume that all of the leptonic h→ ZX events land in a single
bin MZ2 ≈ mX . Defining the total expected number of produced h→ ZZ∗ events as
NZZ∗
prod = σ(pp→ h)× L× Br(h→ ZZ∗ → 4`) (84)
the detector efficiency for dileptons from ZD/a decay can be estimated as
ε`` ≈NZZ∗
detect
NZZ∗prod
, (85)
where NZZ∗
detect is the total expected number of detected h → ZZ∗ events as extracted from
the plots of ATLAS and CMS.37 Therefore, for a given exotic Higgs decay branching ratio,
the expected number of events contributing to the mZ2 distribution is
NZXdetect = ε`` × σ(pp→ h)× L× Br(h→ ZX → 4`)
≈ NZZ∗
detect ×Br(Z → ``)
Br(h→ ZZ∗ → 4`)×[Br(h→ ZX)× Br(X → 2`)
]≈ NZZ∗
detect × 450×[Br(h→ ZX)× Br(X → 2`)
]By placing the above number of events in each mZ2 bin we extract 95% CL bounds on the
quantity in square brackets for different mX > 12 GeV, see Fig. 27.
The bound on Br(h→ ZX)×Br(X → ``) is . 10−4−10−3 for 12 GeV . mX . 34 GeV
and ` = e, µ. Using Fig. 13 we see that this already corresponds to Br(h→ ZZD) . 2×10−3,
which represents a new direct constraint on dark photons by the LHC, see Fig. 12. This limit
can be optimized with a dedicated analysis, which would make LHC measurements the most
sensitive probe of dark vector kinetic mixing in the mass range 10 GeV . mZD . mh/2.
The situation is more ambiguous for pseudoscalars. Their branching ratios are more
model-dependent in general, and their Yukawa couplings usually imply that a → ττ is
37 Due to the mZ2 > 12 GeV requirement this may slightly underestimate the efficiency. There may also
be small differences in isolation for leptonic vector vs pseudoscalar decay. However, our method suffices
for a conservative estimate of constraints.
109
CMS 20 + 5 fb-1
ATLAS 20 + 5 fb-1
15 20 25 30
0
5
10
15
mX
BrHh®ZXL´BrHX®
{{L´104
CMS 20 + 5 fb-1
mX = 23 GeV
BrHh ® Z XL ´ BrHX ® {{L = 5 ´ 10-4
ZZ, ZΓ, Z+X
SM h ® ZZ*h ® ZX
15 20 25 30 350
2
4
6
8
10
MZ2
Events�1GeV
FIG. 27: Left: 95% C.L. exclusion limit on Br(h → ZX)×Br(X → ``) for X = ZD, a, extracted
from the SM h → 4` searches (` = e, µ) assuming SM Higgs production rate and ΓX � 1 GeV.
(The lighter dashed lines indicate the expected limit. The large fluctuations in the observed limit
are a consequence of low statistics in each bin.) Right: The CMS distribution of mZ2 from [176],
overlaid with a 23 GeV h→ ZX → 4` signal.
enormously preferred over e, µ. Typical branching ratios to 4` (` = e, µ) are 10−4 − 10−3,
depending on the pseudoscalar mass. Bounds for X → ττ could also be derived from
the leptonic h → ZZ∗ searches but would be much weaker. Nevertheless this may be the
preferred discovery channel for 2HDM+S and NMSSM type models, where Br(h → Za)
could easily be 10% and Br(a→ ττ) is generally O(0.05–1), see §1.3.2.
10.4. Proposals for New Searches at the LHC
For ma,ZD > 12 GeV it seems likely that LHC14 searches inspired by h → ZZ∗ will
constrain h → Za in the a → 2τ modes, while LHC7+8 already gives significant direct
bounds to h → ZZD → 4`. A Z + lepton-jet search would be able to set strong limits
in particular for very light ZD. Care must be taken to correctly account for challenging
quarkonium backgrounds. Identifying promising search strategies will be the subject of
future work.
110
11. h→ ZDZD → 4`
11.1. Theoretical Motivation
Similarly to the discussion in the previous section, two classes of models can give a Higgs
to four-lepton signature, with two pairs of electrons and/or muons reconstructing the same
resonance:
• As discussed in §1.3.5, models with an additional U(1)D gauge group may lead to the
h → ZDZD decay, followed by ZD → `+`−. In the minimal model, the dark U(1)D is
broken by a dark scalar that does not mix with the SM Higgs. Then the kinetic mixing
operator involving the hypercharge gauge field Bµ and the ZµD field leads to only a
small branching ratio of the Higgs to two ZD gauge bosons, since it is suppressed by
the fourth power of the kinetic mixing parameter ε in Eq. (38). Much larger branching
ratios can be obtained by introducing a mixing term between the scalar that breaks the
U(1)D symmetry and the Higgs of the SM: ζ|S|2|H|2. In these models, even ζ ∼ 10−2
can lead to branching ratios for h → ZDZD as large as ∼ 10% in certain regions of
parameter space (see left panel of Fig. 15). Furthermore, more extended Higgs sectors
can also lead to sizable branching ratios. In particular, in [335] it has been shown that
Br(h→ ZDZD) ∼ 10% is possible in 2HDM+S models where the SM singlet and one
of the two Higgs doublets is charged under U(1)D.
• Many hidden valley models [31, 134] (see §1.3.10), with either fundamental or com-
posite spin-one bosons, can lead to the same final state.
• Models predicting a sizable branching ratio for h→ aa, where a is CP-odd scalar, can
also lead to the 4` signature. As presented in §1.3.2, such pseudoscalars can arise in
2HDM+S models, as for example in the approximately R-symmetric NMSSM scenarios
(see §1.3.7). However, as shown in the figures of §1.3.2, if the pseudoscalar is above
the tau threshold, it will preferentially decay into two taus, two gluons, or two quarks.
More specifically, for ma > 2mτ , Br(a → `+`−)/Br(a → ττ) ∼ m2`/m
2τ ∼ 3 × 10−3
(8 × 10−8) for ` = µ (e). For this reason, in the discussion of §. 11.3 below for the
collider constraints on the 4` signature, we will focus on models with dark gauge
bosons. Searches that exploit the more dominant 4τ and 2τ2µ decay modes of the
111
pseudoscalar pair are discussed in §6.
11.2. Existing Collider Studies
The authors of [166] investigate the feasibility of probing h → ZDZD → 4` at Tevatron
and at the LHC. In particular, they perform an estimation of the reach at the 14 TeV
LHC for several benchmark scenarios: the most interesting for us are the scenarios “A” and
“B” with mh = 120 GeV and mZD = 5 (50) GeV, respectively. They show that there are
very good prospects for detecting this Higgs decay mode, even for small Higgs branching
ratios. In particular, they focus on a Higgs produced in gluon fusion followed by the decay
h → ZDZD → e+e−µ+µ−. For Br(h → ZDZD) ∼ O(1), basic cuts on the pT and η of the
leptons, and the requirement that the 4-lepton invariant mass is close to mh, are sufficient to
lead to S/B ∼ 104 (103) (with S ∼ hundreds (tens) of fb in the case of mZD = 5 (50) GeV).
Here B is simply given by the leading diboson background. Additionally, they comment
on the fact that the reach can be improved further by vetoing events with opposite sign,
same-flavor (OSSF) lepton pairs reconstructing the Z resonance.
Furthermore, Ref. [336] shows that a light Higgs boson could have been discovered sooner
in h→ ZDZD → 4` than in the traditional decay modes, γγ, ττ , with the 7 TeV LHC data.
In particular, the authors claim that, even for Br(h → ZDZD) ∼ O(1%), one could have
expected 5 events with the first fb−1 of 7 TeV LHC data.
11.3. Existing Experimental Searches and Limits
Searches for h→ aa→ 4µ were performed by the CMS collaboration with 5 fb−1 of data
at√s = 7 TeV [286] and 20 fb−1 at
√s = 8 TeV [337]. For these searches, a refers to a
spin-0 boson with a mass between 250 MeV and 2mτ . Differences in the acceptance between
this signal and h→ ZDZD → 4µ should be modest for this range of boson masses, and the
limits from these searches at CMS are directly applicable. The 8 TeV search [337] is more
sensitive and results in a limit Br (h→ ZDZD → 4µ) < 4.7× 10−5 for mh = 125 GeV and
250 MeV < mZD < 2mτ .
For the mass range 5 GeV < mZD < mh/2, limits can be obtained from SM Higgs searches
as well as from a plot reported as part of a ZZ cross section measurement. To estimate
112
limits on exotic Higgs decays to four leptons, we use MadGraph to generate Higgs decays
to dark photons, h → ZDZD, followed by ZD → `+`−, using FeynRules [338] to construct
the dark photon model of §1.3.5. Gluon fusion signal events are generated in MadGraph 5
and matched up to one jet, with showering in Pythia.
We begin by considering the SM h → ZZ∗ analyses, which are conducted with the full
7+8 TeV datasets in both experiments. The CMS search [176] requires four isolated leptons
within kinematic acceptance, forming two OSSF pairs. The invariant mass of the OSSF pair
that minimizes |m`` −mZ | is denoted m1, while the remaining OSSF pair invariant mass is
denoted m2. The pair invariant masses must satisfy
This reduces the overall acceptance for the BSM signal, leading to weaker limits than those
from CMS (as both experiments observed 4 total events in the signal region, and as ATLAS
does not report flavor information for these events).
At low masses, the best limits are found from control regions in the ATLAS ZZ cross
section measurement with 20 fb−1 of 8 TeV data [339]. Here, events are again required to
have exactly four leptons, which can be paired into two OSSF pairs. Now when there is a
choice of possible OSSF pairings, the assignment which minimizes |m1 −mZ | + |m2 −mZ |
is chosen. This still has some probability of mis-pairing h → ZDZD events, as can be seen
in Fig. 29. The invariant mass of the lepton pair with higher pT is assigned to be m1. Note
that, unlike the SM h→ ZZ∗ analyses, there is no restriction on the invariant mass of the
four leptons.
We now set limits by defining a signal region for each mass, mZD−2 GeV < m1 < mZD+2
GeV, mZD − 2 GeV < m2 < mZD + 2 GeV. Lepton efficiencies are modeled with a pT -
dependent parameterization for electrons [341, 342] and a flat efficiency for muons, and
validated against the fiducial acceptances for ZZ events quoted in [339]. At most one event
116
is observed in each 4 GeV × 4 GeV signal bin. Treating any observed event in the signal
region as signal, we obtain 95% CL limits as before.
Fig. 30 shows the resulting limits (along with those from CMS’s h → ZZ∗ search), of
order 10−3, on Higgs branching fractions to dark vector bosons that further decay to lepton
pairs. These limits, while impressive, are easy to improve at low masses by simply looking
for OSSF pairs which minimize |m1 − m2|, instead of a distance from the Z peak. As
backgrounds are already zero for most bins, improving signal acceptance is the most likely
to improve reach.
12. h→ γγγ + E/T
We consider here the signature h → γ + E/T . This signature can be usefully represented
through the decay of the Higgs into two neutral fermions, h→ χ1χ2, followed by the decay
χ2 → γχ1.
12.1. Theoretical Motivations
While our focus here is on decays to BSM particles, it is worthwhile to observe that
the signature h → γ + E/T arises as a rare decay in the SM, through the loop-induced
h → γZ, followed by Z → νν. The SM branching fraction is thus Br(h → γ + νν)|SM =
1.54 × 10−3 × 0.20 = 3.08 × 10−4 [343]. Searches for potential enhancements in h → γZ
are sensitive to the potential presence of new physics running in the loop, making this rare
Higgs decay signature one of interest for several reasons. The decay h→ γZ implies specific
kinematics for the photon and missing energy, however, which do not hold in more general
models.
One class of models that gives rise to a h→ γ+E/T signature are those with very low-scale
supersymmetry breaking [344]. Here the Higgs decays into a gravitino and a neutralino that
is dominantly bino, h→ GB, followed by the prompt decay B → γG [50]. As the gravitino
is effectively massless, this model is parameterized by one mass mB. This mass should lie
in the range mh/2 < mB < mh to obtain a large branching ratio to h → γ + E/T , as for
mh/2 > mB, the decay h→ BB will dominate, leading to a h→ 2γ + E/T signature.
This signature can also be realized in the PQ-limit of the NMSSM (see §1.3.8). Here
117
the lighter fermion χ1 is dominantly singlino, and the heavier fermion is dominantly bino.
The mass splitting between the two fermions is now much more free. However, in the
PQ-symmetric limit, a light singlino is always accompanied by a light scalar s, and for
the loop-induced branching fraction Br(χ2 → χ1γ) to be sizable, the tree level decays
Br(χ2 → s(∗)χ1 → ffχ1) must be phase-space suppressed. Thus one generically expects
mass splittings between the two neutralino species of no more than 10-20 GeV for the rate
into h → γ + E/T to be appreciable. Outside the PQ-symmetric limit of the NMSSM, or in
other extensions of the MSSM [345], special parameter cancellations are required to obtain
substantial branching fraction for the radiative decay χ2 → γχ1.
A more bottom-up approach extends the SM by two Majorana fermions, χ2 and χ1, with
a dipole coupling
δL =1
µχ2σµνB
µνχ1. (88)
Note that the presence of the hypercharge field strength B would predict a Z + E/T signal
as well, if phase space allowed it; however, in many UV completions of the dipole operator,
the mass-splitting between the fermionic states arises due to some symmetry breaking which
makes it challenging to realize mχ2 −mχ1 & mZ , and the Z mode will typically be highly
suppressed. The simplified model is then characterized by two parameters m1 and m2, where
m1 < m2 and m1 +m2 < mh.
Finally, the γ+E/T signature also appears as a subleading decay mode in models of Higgs
decay to right-handed neutrinos N [346]. Here the signature arises from h→ NN , followed
by the decay of N → γν on one side of the event and N → ννν on the other. In the
realization of [346], both of these N decay modes are highly subdominant, and the photonic
decay may be displaced.
12.2. Existing Collider Studies
An LHC study was carried out at parton level in [50]. This study targets Higgs bosons
produced in gluon fusion and estimates that 20 fb−1 of 8 TeV data would allow 95% CL
sensitivity to branching fractions ranging between Br(h → γ + E/T ) < 0.002 for mχ2 =
120 GeV, and Br(h → γ + E/T ) < 0.010 for mχ2 = 60 GeV. These results are based on
selection criteria that are not obviously compatible with current LHC triggers, however, as
118
the selection of Ref. [50] requires
45 GeV < pTγ <mh
2(89)
and no other triggerable objects. Current monophoton triggers require pT,γ > 80 GeV,
although trigger cuts for CMS parked data are more relaxed, pT,γ > 30 GeV and E/T >
25 GeV for central photons, and therefore could be relevant for this decay channel.
Replacing the cut on photon pT with one on the transverse mass of the photon and
the missing momentum gives a good separation between signal and backgrounds. Trigger
thresholds ensure that the dominant contribution to the reach comes from the high-pT
tail of the Higgs production spectrum, where the Higgs recoils against one or more hard
ISR jets. Depending on the mass difference between χ1 and χ2 and the analysis threshold
achieved in parked monophoton +E/T triggers, the best signal acceptance may be achieved
in monojet+E/T -triggered events rather than monophoton+E/T -triggered events.
12.3. Existing Experimental Searches and Limits
In (N)MSSM realizations of the nonresonant signature, there are indirect limits on the
Higgs branching fraction into neutralino-gravitino from electroweak-ino searches at Tevatron
and at LHC (see also the nonresonant 2γ + E/T signature in §13, where similar considera-
tions apply). In the case of the neutralino-gravitino realization, the lightest neutralino χ1
must have some Higgsino component in order for the coupling hχ1G to be present. In the
neutralino-singlino realization, the heavier fermion χ2 is typically dominantly B, with χ1
dominantly singlino, and the vertex hχ2χ1 again proceeds through the Higgsino component
of χ2. In both scenarios the non-zero Higgsino component implies the bino-like state should
be produced directly at hadron colliders via Drell-Yan [72], which may or may not lead to
constraints depending on the ensuing decay modes of the bino. While it is of interest to
work out these indirect limits, the surviving parameter space is multidimensional, and in
more general models, where the coupling hn2n1 arises from a dimension-five Higgs portal
coupling, the new neutral fermions do not need to have tree level couplings to the Z boson,
and no such indirect limit applies.
Very few existing collider searches place any limits on Br(h → γ + E/T ). Searches for
a hard photon plus E/T , designed to pick up invisible particles recoiling aginst a hard ISR
119
photon [347–349], target very different kinematic configurations and are not constraining.
Similar conclusions apply to the Zγ, Z → νν cross-section measurements [350, 351], which
also target high-pT photons recoiling against E/T .
Searches for supersymmetry in final states with γ + ` + E/T+jets at the LHC [352, 353]
and the Tevatron [354] can be sensitive to Wh associated production when the W decays
leptonically. Acceptance for the Higgs signal in these supersymmetry searches is small,
due to the hardness demanded of both the γ and the E/T . No limit is placed by the LHC
Wh searches in any part of the m1-m2 simplified model parameter space. The Tevatron
searches likewise place no limits, partially due (particularly for large m2 − m1) to a 1σ
excess of observed events relative to expectation. This quick limit check assumes 100%
photon efficiency; incorporating realistic photon efficiency would further weaken the search.
The general CDF search for anomalous γ +E/T+at least one jet also does not constrain the
Higgs branching fraction [355].
CMS’ supersymmetry search in the γ + E/T+jets final state [356] comes closer to being
constraining; again, no limits are placed anywhere in the m1-m2 simplified model parameter
space, but as before this lack of constraint is partially due to a 1.3σ excess of events observed
over background expectation (assuming 100% photon efficiency). An updated search in the
same final state [357] with 4.04 fb−1 of 8 TeV data requires all events to have HT > 450 GeV,
giving punishingly small signal effiency. Despite the harshness of this cut, this analysis is
beginning to gain sensitivity to the γ + E/T decay mode, as shown in Fig. 31. The reported
limits from [357] are difficult to recast due to the existence of signal contamination in a
region E/T < 100 GeV used to model the dominant QCD background. The light 125 GeV
Higgs contributes proportionately more to the control region E/T < 100 GeV than do the pair
produced neutralinos with mass 375 GeV for which the background predictions are shown.
The limits found by recasting the analysis for a light Higgs are likely overconservative to an
extent that is difficult to estimate. In Fig. 31 we show the result of performing this simple
recast. The signal region is divided into multiple exclusive bins in E/T , with background
predictions as reported for the pair-produced neutralinos. We place limits by combining the
limits from each individual bin using a Bayesian algorithm with flat priors, and marginalize
over background uncertainty according to a lognormal distribution. With perfect photon
efficiency, the 95% CL limits obtained on Br(h→ γ +E/T ) is approximately unity in a large
range of parameter space, suggesting that an analysis more tailored to the signal kinematics
120
40 GeV
20 GeV
0 GeV
0 20 40 60 80 100 1200.5
1.0
1.5
2.0
2.5
mΧ2qHGeVL
Brk Hh
®Χ
1Χ
2L
FIG. 31: Approximate 95% C.L. upper limit on (σ/σSM ) × Br (h→ χ1χ2 → γ + E/T ) from the
results of Ref. [357], for mχ1 = (0 GeV, 20 GeV, 40 GeV) < mχ2 . Solid lines correspond to 100%
photon efficiency, and dashed lines to a (flat) 80% photon efficiency.
could place meaningful limits on the branching fraction for this channel.
As with all semi-invisible signals, collider reach could be extended by forming the trans-
verse mass of the visible decay product(s), here the photon, with the missing transverse
momentum vector, and requiring this to be bounded from above as consistent with produc-
tion from an initial resonance. Much better sensitivity could be achieved if the prohibitively
hard cut on HT could be relaxed. This HT cut is necessitated by the γ+HT trigger used to
select the data in the current analysis, and is not suited well to the study of the relatively
low-pT Higgs events. Somewhat better signal acceptance is realized for the monophoton+E/T
triggers in current use for dark matter searches, though the degree of improvement depends
on the spectrum; again, monojet+E/T triggers may provide better sensitivity.
13. h→ 2γγγ + E/T
In this section we consider the decay h → 2γ + E/T . This signature can be realized in
several ways.
• First, consider the non-resonant signature where the photons come from opposite sides
of the initial two-body decay, h→ XX, followed by X → γY on each side of the event
121
with Y a detector-stable, neutral particle.
• Second is the case where the photons reconstruct an intermediate resonance, h→ XX,
with X → γγ on one side and X → invisible on the other.
• The last decay topology we consider involves the initial decay h → XY , followed by
X → Y φ, φ→ γγ with Y again appearing as missing energy in the detector.
These different cases may arise in different theoretical models, and require related but dis-
tinct strategies to observe at colliders, as we discuss below.
13.1. Theoretical Motivation
13.1.1. Non-Resonant
The non-resonant decay of the Higgs boson to two photons and missing energy may be
realized in several theoretical scenarios.
As a first example, consider gauge-mediated supersymmetry-breaking models. Here the
lightest neutralino is mainly bino, and decays via χ01 → γG. Minimal models of gauge
mediation make it difficult to obtain a bino with mB < mh/2 while keeping winos sufficiently
heavy to satisfy LEP bounds on the charginos as well as gluinos sufficiently heavy to avoid
LHC constraints. However, more general models of gauge mediation [358] can allow this
spectrum to be realized [72].
Another realization of the non-resonant 2γ + E/T signature may be obtained in the PQ
limit of the NMSSM (see §1.3.7 and §1.3.8 for more details), where a light singlino s replaces
the gravitino. In this case the photonic signature is realized through a loop-induced dipole
coupling B†σµνBµν s. There are typically several other decay modes available to the B in
these NMSSM models, in particular
B → Z(∗)s, B → a(∗)s, B → s(∗)s, (90)
where a, s are light, dominantly singlet CP-odd and CP-even scalars. The radiative decay
B → γs is typically significantly subdominant to the tree-level decays. The 2γ+E/T signature
is thus typically small compared to other exotic decay modes in the PQ NMSSM.
122
More generally, this signature may be realized by having two new (Majorana) fermions
χ1 and χ2, with a dipole coupling
δL =1
µχ†2σµνB
µνχ1 (91)
and a dimension five Higgs portal coupling c22|H|2(χ2χ2 +χ†2χ†2). In this case, both mχ1 and
mχ2 are parameters of the model. It is natural to extend this simple model to include in
addition off-diagonal couplings c12|H|2(χ2χ1 + χ†2χ†1) and couplings of the Higgs directly to
the lighter of the two new fermions, c11|H|2(χ1χ1 + χ†1χ†1). This generic model would then
also yield h → 1γ + E/T and h → E/T signatures with relative branching fractions uniquely
determined by the cij. Previous study of this topology in the MSSM has been performed in
[72] and, for the heavier MSSM Higgses, in [359]; see also [51]
13.1.2. Resonant
The 2γ+E/T final state can also occur for the decay chain h→ aa, with one intermediate
state decaying to photons, a → γγ, and the other decaying invisibly, a → inv. This can
be simply realized in a bottom-up fashion by introducing a renormalizable Higgs portal
interaction leading to a coupling of a to h, λ |H|2 a2, and also coupling a to photons and to
a neutral, detector-stable particle χ via, e.g.,
α
4πMaF µνFµν +
∂µa
M ′ χγµγ5χ. (92)
M and M ′ are the scales of the two dimension-five operators, and we have assumed that a
is a real pseudoscalar and that χ is a Dirac fermion for definiteness. For some regions of
parameter space, a → γγ and a → χχ can have comparable branching fractions, making
h→ 2γ +E/T an important final state. Another possibility arises from the decay chain h→
χ1χ2 → aχ1χ1, where a decays via the first dimension-five operator and χ1 is stable. Note,
though these two decay topologies can be achieved in the R- and PQ-limits in the NMSSM
(see §1.3.7 and §1.3.8), the branching fraction of a→ γγ tends to be small. Alternatively, a
may be light enough so that a→ ff is kinematically suppressed, in which case the lifetime
is so long that a would decay outside the detector. More general models may give a larger
a→ γγ coupling than the NMSSM.
Unlike the non-resonant case, the resonant signature has the useful additional handle
that the two photons should reconstruct ma, improving the search prospects. Additionally,
123
as ma is decreased and the intermediate particles become more boosted, a larger fraction
of the photon pairs will fail isolation cuts. For mh = 125 GeV, this becomes important for
ma . few GeV. In this case, the signal would have some overlap with that from h→ γ+E/T
considered in §12 [315].38
This simplified model can be trivially generalized to the case that the Higgs decays to two
distinct states, a1 and a2, with a1 → γγ and a2 → inv. This can proceed through a dimension-
four Higgs portal interaction, λ12|H|2a1a2, if a1 couples to photons while a2 decays invisibly.
This decay mode can dominate over h → inv. or h → 4γ if λ12 � λ11,22 where λ11,22 are
the coupling constants of the other allowed Higgs portal interactions, λ11|H|2a21 +λ22|H|2a2
2.
While, in this resonant case, we limit our study to the situation ma1 ' ma2 ≡ ma, the two
intermediate particles having different masses is a well-motivated possibility.
13.1.3. Cascade
The h→ 2γ + E/T decay can proceed through h→ χ1χ2, with χ2 → sχ1, s→ γγ if χ1 is
neutral and stable on detector scales. It is easy to write down a simple model that gives rise
to this decay chain. We can couple (Majorana) fermions χ1 and χ2 to the Higgs through a
dimension-five Higgs portal coupling as in the non-resonant case above, c12|H|2(χ2χ1+χ†2χ†1),
as well as to the scalar s through a Yukawa interaction, y12s(χ2χ1 + χ†2χ†1). Furthermore, s
can decay to two photons through the dimension-five operator sFµνFµν .39
13.2. Existing Experimental Searches and Limits
In (N)MSSM realizations of the non-resonant signature, there are indirect limits on the
Higgs branching fraction into neutralinos from general electroweak-ino searches at the Teva-
tron and at the LHC. These limits arise because the lightest neutralino χ01 must have some
Higgsino component in order for the coupling hχ01χ
01 to be present. Because of this non-zero
38 In the ma � mh regime, the relationship between the h→ 2γ+E/T and h→ γ+E/T signals parallels that
between h→ 2γ and h→ 4γ. See §9 for further details.39 The sFµνF
µν operator could arise through mixing between s and h, see for example §1.3.1, although that
would lead to a very suppressed h→ 2γ + E/T branching ratio compared to final states like bb+ E/T . For
2γ + E/T to be dominant, the sFµνFµν operator would have to be generated by a direct coupling of s to
electrically-charged matter, e.g. (heavy) vector-like leptons. For a similar model, see §8.
124
20 GeV
30 GeV
40 GeV50 GeV
60 GeV
0 10 20 30 40 50
0.02
0.05
0.10
0.20
0.50
1.00
mΧ1kHGeVL
Brk Hh
®Χ
2Χ
2L
FIG. 32: Approximate 95% C.L. upper limit on (σ/σSM ) × Br (h→ χ2χ2 → 2γ + E/T ) from the
results of Ref. [357], for multiple values of mχ2 as indicated by the text labeling the different curves.
Solid lines correspond to 100% photon efficiency, and dashed lines to a (flat) 80% photon efficiency.
Higgsino component, the lightest neutralino couples to the Z and can be produced directly
at hadron colliders via Drell-Yan. Model-dependent indirect limits on Higgs branching frac-
tions arising from Drell-Yan direct production are nontrivial [72] and an interesting topic
of study, but in the present work we confine ourselves to considering (model-independent)
direct limits, and make no assumptions about other production modes for the BSM states.
In general (non-MSSM) models, where the coupling hχ2χ2 arises from a dimension five Higgs
portal coupling, the new neutral fermion χ2 does not need to have tree-level couplings to
the Z boson, and those indirect limits do not apply.
In GMSB realizations of the non-resonant signal, sufficiently high SUSY-breaking scales
lead to a macroscopic decay length for the neutralino. This can also occur in the general
Higgs portal simplified model, for sufficiently large dipole suppression scales µ in the decay
vertex of Eq. (91). In such cases, non-pointing photon searches may be motivated or neces-
sary. Displaced signatures are beyond the scope of the present work, but are an interesting
and natural avenue for future exploration.
GMSB searches at the LHC have good prospects for discovering or excluding exotic Higgs
decays into 2γ + E/T , in both the resonant and non-resonant scenarios. The ATLAS search
for 2γ+E/T using 7 TeV data [360] has some sensitivity, setting limits of . 15% on the exotic
Higgs branching fraction over much of the parameter space. The more recent CMS study
125
0 10 20 30 40 50 60
0.02
0.05
0.10
0.20
0.50
1.00
ma @GeVD
HΣ�Σ
SM
L´B
rHh®
ΓΓ+
inv.L
0 10 20 30 40 50 60
0.02
0.05
0.10
0.20
0.50
1.00
ma @GeVD
HΣ�Σ
SM
L´B
rHh®
ΓΓ+
inv.L
FIG. 33: Approximate 95% C.L. upper limit on (σ/σSM ) × Br (h→ 2γ + E/T ) from the 2γ + E/T
search in [357]. The solid lines correspond to 100% photon efficiency, and the dashed lines to a
(flat) 80% photon efficiency. Left: Resonant case, where h → aa, one a decays to γγ and the
other decays invisibly. Right: Cascade case, where h → χ1χ2, χ2 → sχ1, s → γγ. Here mχ1 = 0
and mχ2 = 60 GeV (although the limit is insensitive to the particular value of mχ2 as long as it is
kinematically allowed).
using 4.04 fb−1 of 8 TeV data [357] sets the current best limits. This search selects events
with at least two photons and at least one central jet, and bins events in 5 exclusive E/T bins
beginning from a minimum of 50 GeV. We show the reach of this search in the resonant
and non-resonant cases in Figs. 32 and 33 (left), as a function of mχ1 in the non-resonant
topology and ma in the resonant topology. In Fig. 33 (right), we show the reach in the
case of the cascade topology as a function of ms, setting mχ1 = 0 and mχ2 = 60 GeV. We
find that the limit obtained in this case is not very sensitive to the value of mχ2 = 60 GeV
chosen. In all three topologies the Br(h→ 2γ +E/T ) can be constrained at the level of a few
percent over much of the parameter space. Higgs signal events are generated in MadGraph
with showering in Pythia, and jet clustering is done with FastJet. Gluon fusion is matched
out to one jet, and cross-sections for both gluon fusion and vector boson fusion processes are
set to the values determined by the LHC Higgs Working Group [302]. VBF production is
responsible for 20-25% of the signal. To obtain limits we combine individual 95% CL limits
from each of the 5 E/T bins according to a Bayesian algorithm with flat priors, marginalizing
over the background uncertainty according to a log-normal distribution.
Since searches using only 4 fb−1 of 8 TeV data and optimized for other signatures are
126
already able to place limits as stringent as O(5%) on the Higgs branching fraction into this
mode, 2γ+E/T is a good candidate for searches in the near future. The reach could be easily
extended by requiring the transverse mass of the photons and E/T to be bounded from above,
as consistent with resonant origin from the 125 GeV Higgs. In the resonant case, looking
for a peak in the γγ spectrum could offer another useful handle.
14. h→ 4 ISOLATED LEPTONS + E/T
Exotic Higgs decays into multiple charged leptons together with missing energy are less
frequently motivated by top-down model building than (e.g.) h→ aa cascade decays, but on
the other hand, they offer excellent discovery potential at the LHC, as we will demonstrate
in this and following sections.
There is some overlap between the theoretical motivations and decay topologies for dif-
ferent h→≥ 2 charged leptons +E/T +X signatures. Here we briefly discuss all the cases we
consider in this document before treating the 4`+ E/T case in detail.
Depending on the specific model under consideration, the characteristic predictions for
leptonic final states can be very different. Exotic Higgs decays h→ X1X2 (where X1,2 may
or may not be distinct species) can be divided into two main classes of topologies:
1. `+`− + E/T , which involves the topologies:
• I: X1 → non-leptonic+E/T , X2 → `+`− + E/T
• II: X1 → non-leptonic+E/T , X2 → `+`−
where the the non-leptonic part is typically either nothing (i.e., X1 stable and invisible)
or hadronic (i.e., X1 → soft jets+E/T );40 and
2. 2× `+`− + E/T , which can be achieved via the topology
• III: X1 → `+`− + E/T , X2 → `+`− + E/T .
• IV: X1 → `+`−, X2 → `+`− + E/T .
40 Charged X’s each decaying to `+ E/T are highly constrained, and not considered here.
127
Further, the cascade decays of X2 in topologies I and III may either be three-body, or
they may involve an on-shell intermediate state so that the leptons reconstruct a resonance.
Depending on the mass of this resonance, and similarly on the mass of the X2 resonance in
topologies II and IV, the leptons may be either isolated or collimated.
This gives us a plethora of experimental signatures, all of which present interesting targets
with the existing LHC dataset. We discuss theoretical models and experimental prospects
for these leptonic signatures here and in the following two sections. In the current section
we discuss final states with four isolated leptons plus missing energy; in §15 we discuss final
states with two isolated leptons plus missing energy; in §16 and §17 we consider final states
that include one or two lepton-jets, respectively; decays to leptons without E/T are discussed
in §10 and §11.
14.1. Theoretical Motivation
Several classes of models can give rise to Higgs decays to 4 isolated leptons+E/T . First,
consider models with weak-scale neutral states that have non-vanishing couplings to the
Z boson, such as exotic neutrinos or neutralinos. In this case, leptons can arise from the
three-body decay of one neutral fermion χ2 to a lighter one χ1 through an off-shell Z boson,
appearing as an opposite-sign, same flavor pair. The 4`+E/T signal then arises from cascades
of the form h → χ2χ2 → χ1Z∗χ1Z
∗ with both Z∗ leptonic. In fourth-generation neutrino
models, χ2, χ1 are the two Majorana-split halves of a Dirac neutrino state; in MSSM-like
realizations, χ2, χ1 are neutralinos. The branching fraction into 4`+ E/T is small compared
to the total branching fraction into χ2χ2: Br(h → 4` + E/T )/Br(h → χ2χ2) = Br(Z →
``)2 ≈ 0.011 (including τs). Despite the small relative branching fraction, we will see that
the 4` + E/T final state is typically more constraining than final states with fewer leptons,
due to the low backgrounds for multi-leptonic final states.
Hidden sectors with a kinetically mixed dark vector boson ZD can also realize this decay
chain [31, 232]. For instance, a hidden sector with meson-like pseudoscalar states Kv, πv,
may have a spectrum such that the heavier meson may only decay via Kv → Z∗Dπv → ffπv,
and the lighter meson πv is collider-stable. The width for this Kv decay scales like
ΓKv ≈ αDαEMε2
15 cos θ2W
(mKv −mπv)5
m4ZD
, (93)
128
where ε is the kinetic mixing between hypercharge and the dark vector boson (see §1.3.5).
The Kv meson decay can be prompt provided the ratio of the dark meson mass splitting to
the dark photon mass, (mKv−mπv)/mZD , is not particularly small. The branching fractions
into leptonic final states are much larger here than in the case where the three-body decay
is mediated by a virtual Z. For a dark vector with mZD > 2mb & 10 GeV, the branching
fraction into leptonic final states (including taus) is Br(ZD → leptonic) ≈ 45%, as discussed
in §1.3.5.
Another realization of this type of decay chain with an off-shell kinetically mixed dark
photon occurs in supersymmetric hidden sectors, with one or more hidden neutralinos. In
this case the Higgs cascade decay could begin with a Higgs decay to bino-like neutralinos B,
which in turn decay via B → Z∗Dχ01, where χ0
1 is a hidden sector neutralino [51, 148, 361].
If the dark photon is sufficiently light, the decay Kv → ZDπv → ``πv can be allowed,
and the leptons reconstruct a resonance at m`` = mZD . In the PQ-symmetric limit of the
NMSSM, light (pseudo)scalars in the spectrum similarly enable the on-shell decay χ2 →
s(a)χ1 → ``χ1. However, in the NMSSM, the branching fractions to light leptons are
suppressed by small Yukawa couplings, and Br(h → 4µ + E/T ) is cripplingly small unless
the scalar is below the τ threshhold, ms(a) < 2mτ . When the scalar is this light, it is often
produced with pT,s � ms, leading to collimated muons, but this is spectrum-dependent.
Collimated lepton pairs (lepton-jets) are discussed in §16 and §17.
In models with a nontrivial flavor structure, flavor-violating decays of the form h→ χχ→
4`+ 2ν can occur. A familiar example is Higgs decay into R-parity violating neutralinos χ1,
where χ1 decays through the leptonic LiLjek operator. In this case the two charged leptons
from the decay χ1 → `′`ν need no longer necessarily form same-flavor pairs.
Finally, another realization of the same final state occurs when the Higgs decays into
two heavy neutrinos N , which then each decay through N → W ∗` → ν`′` [129]. Similar
phenomena and final states can arise in scotogenic models [362, 363].
14.2. Existing Experimental Searches and Limits
Several LHC searches give interesting bounds on the exotic decay h→ 4`+E/T . The best
bounds when the leptons are non-resonant come from 8 TeV LHC multi-lepton searches. In
order to highlight the strong dependence on the exotic spectrum, we will present bounds for
129
two benchmark models where h→ χ2χ2 and χ2 → χ1Z∗:
• An “optimistic” benchmark scenario with relatively large mass splitting between χ2
and χ1, with M2 = 55 GeV and M1 = 20 GeV. Generally, models of this type are
allowed by the LEP precision measurement of the Z width, as long as the coupling
of the Z boson to χ1χ2 is smaller than ∼ 0.05.41 Even for couplings O(0.01), the
decay χ2 → ``χ1 is prompt. In general the Drell-Yan production of χ2χ2 will yield
an additional and model-dependent contribution to the leptons+E/T signature. For
simplicity, throughout our analysis, we will always assume that the Z coupling to
χ2χ2 is sufficiently small that the Drell-Yan contribution is much smaller than the
contribution coming from Higgs decay.
• A “pessimistic” benchmark scenario with a smaller mass splitting, M2 = 55 GeV and
M1 = 35 GeV. This particular parameter point is consistent with LEP data when
χ2, χ1 have the Z couplings of fourth-generation neutrinos [127]. The relatively small
mass difference between the exotic final states renders the final state leptons softer
and makes the benchmark more challenging at the LHC.
In both cases we take
Br(χ2 → `+`−χ1) = Br(Z(∗) → `+`−). (94)
For Higgs bosons produced in gluon fusion and assuming a reference 10% branching ratio
for h→ χ2χ2, the initial signal cross section for
pp→ h→ χ2χ2 → 4`χ1χ1 (95)
is approximately 10 fb, giving already ∼ 200 events in the present LHC data set. Below we
will indicate the excellent potential of the LHC to set bounds on the optimistic benchmark
by recasting existent searches in multi-leptons. To indicate the sensitivity of these searches
to the mass splitting between χ1, χ2 we also show that the more pessimistic benchmark,
with its much softer daughter leptons, is as yet unconstrained. Dark photon models, with
larger branching fractions to leptonic final states, face more stringent limits.
41 This number has been found under the assumption gV = gA where gV and gA are the vector and axial-
vector couplings gV Zµχ2γµχ1 and gAZ
µχ2γµγ5χ1, respectively. Similar limits can be found for gV 6= gA.
130
The multilepton analysis strategy pursued by both ATLAS and CMS divides events into
several exclusive bins depending on multiple variables. The variables most notable for our
purposes are: lepton counts N`; OSSF lepton pair invariant masses; and either (1) the value
of E/T and HT (defined as the scalar sum of the transverse energies of all jets passing the
preselection cuts) [296, 364], or (2) the value of ST (the scalar sum of E/T , HT , and the
pT of all isolated leptons) [297, 365], or (3) the value of mT in three-lepton searches [366].
A more inclusive strategy is pursued in [367], which uses only N` and lepton pair invariant
masses to define the several signal regions, while [368] introduces more specialized kinematic
constraints to target specific models of electroweak production. All of these analyses set
limits on models beyond the SM by combining individual limits from all bins, both high-
background and low-background. As reinterpreting multi-lepton searches is highly sensitive
to the details of modeling lepton acceptance, our aim here is principally to demonstrate the
interesting level of sensitivity already available to non-resonant multi-leptonic Higgs decay.
In order to estimate signal efficiency, we generate inclusive Higgs events with at least 2
leptons42 in MadGraph 5, shower them in Pythia, and cluster them in FastJet. We generate
gluon fusion production matched to one jet, VBF, and Wh associated production. The
signal production cross-sections are normalized to the values reported by the LHC Higgs
Working Group [12] (see Table I).
For CMS multilepton analyses, we are able to make a fairly precise approximation of the
signal efficiency by passing signal events through the version of PGS tuned by the Rutgers
theory group [112, 369] to more exactly simulate the CMS detector.43 We employ in addition
the modified b-tagging routines and the correction factors for electron, muon, and hadronic
tau efficiencies as established in [370].
For the ATLAS multilepton analyses, we approximate signal acceptance using the pT -
dependent lepton identification efficiencies quoted in Refs. [341, 342]. Since our signal is
characterized by relatively soft leptons, it is important to note that the electron efficiency
drops below 70% for peT . O(10) GeV while the muon identification efficiency remains high
even for very soft muons (∼ 90% for pµT & 7 GeV).
To set limits we treat each bin as a single Poisson counting experiment, marginalizing over
42 We include taus in the generation of the events. Taus are decayed using the Tauola plugin within Pythia.43 Thanks in particular to M. Park and S. Thomas.
131
background uncertainty according to a log-normal distribution, and combine bins according
to a Bayesian algorithm with flat priors on signal strength. We quote 95% CL upper bounds.
The best limits on the optimistic benchmark come from recasting the 19.6 fb−1 search
performed by CMS in four-lepton final states [367]. This search requires exactly four light
leptons in the final state, forming at least one OSSF pair. Denoting the invariant mass of
the OSSF lepton pair with mass m`` closest to mZ as M``1 and the invariant mass of the
remaining lepton pair as M``2, the events are divided into 9 exclusive categories depending
on whether M``1 and M``2 are below, above, or inside the Z window 90± 15 GeV. The vast
majority of exotic Higgs decays fall in the bin M``1 < 75 GeV, M``2 < 75 GeV. Indeed, this
is the only bin populated by gluon fusion and VBF; Wh associated production is the only
contributing process in the other bins. The combined limit from all populated bins is
Br(h→ χ2χ2) < 11%, (96)
which is also the 95% CL limit set by the single dominant bin. This translates into the
limit Br(h → 4` + E/T ) < 1.2 × 10−3, with ` = (e, µ, τ)44 for dark vectors with Br(ZD →
``) = 3 × 0.15, Br(h → KvKv) < 6.1 × 10−3. We show predicted signal events for this bin
together with the expected and observed number of events in Table VIII. To show the steep
dropoff in signal acceptance when the mass splitting in the cascade decay becomes smaller,
we also show signal predictions in the same bin for the pessimistic benchmark, where the
acceptance in gluon fusion has almost entirely disappeared.
The CMS three- and four-lepton channel search of Ref. [297], done with 9.2 fb−1 of 8 TeV
data, places a similar limit of
Br(h→ χ2χ2) < 14%. (97)
The signal dominantly populates the lowest bin in ST , namely 0 < ST < 300 GeV, for
all lepton multiplicity channels; VBF production also contributes secondarily to the next-
highest bin, 300 GeV < ST < 600 GeV. The bin with the single greatest signal contribution
is that with three identified leptons and one OSSF pair with mass below the Z window.
However, the signal-to-background ratio is better in the bin with the second-largest number
of signal events, namely the bin with four identified leptons and two OSSF pairs below the
Z window, no b’s, and no hadronic taus. This bin dominates the limit combination.
44 Note that this limit translates into Br(h→ 4`+ E/T ) < 5.4× 10−4 considering simply ` = e, µ.
132
Model Mode CMS bin Prediction [367] ATLAS bin Prediction [366]
“Optimistic” gluon fusion 50.4 2.4
(M1 = 20 GeV, VBF 56.2 7.6
M2 = 55 GeV) Wh 2.1 14
total 109 24
“Pessimistic” gluon fusion – 0.6
(M1 = 35 GeV, VBF 2.2 2.2
M2 = 55 GeV) Wh 0.2 3.6
total 2.4 6.4
TABLE VIII: Benchmark predictions for the number of events in the dominant bin (see text) in the
most constraining CMS multi-lepton search [367] (third column) and ATLAS three-lepton search
[366] (fourth column), for the optimistic and pessimistic benchmarks defined in the text, with
Br(h → χ2χ2) = 1 and Br(χ2 → χ1``) = Br(Z → ``). In the CMS bin, 14 events are observed
and 10.4 ± 2.0 are expected. In the ATLAS bin, 41.8 events are excluded at the 2σ level. Signal
expectations are reported separately for gluon fusion, VBF, and associated Wh production.
For the pessimistic benchmark, Ref. [297] limits
σ(pp→ h)
σ(pp→ h)|SMBr(h→ χ2χ2) < 1.04, (98)
or Br(h → 4` + E/T ) < 0.011. The reach is almost entirely from VBF production, with
several bins contributing significantly to the limit.
The CMS search of Ref. [296] uses the same data set as Ref. [297] but bins events in E/T
and HT instead of in ST , and sets comparable limits. Finally, the CMS searches performed
in Ref. [368] use kinematic discriminants which are tailored to the electroweak production
of heavy states, and are not sensitive to the kinematics of our exotic Higgs decay signal.
ATLAS multilepton searches [365, 366] are less sensitive than the CMS searches we have
just discussed, mainly because of the missing energy requirement (at least 50 GeV in all the
signal regions). In particular, the most sensitive search is the three-lepton search of [366]
performed with 20.7 fb−1 of 8 TeV data. The most constraining bin is the so-called SRnoZa
that requires E/T > 50 GeV and all OSSF lepton pairs to have a invariant mass below 60 GeV.
As shown in Table VIII, the main contribution to this bin comes from a Higgs produced in
133
association with a W boson. Assuming Br(h→ χ2χ2) = 1, the optimistic benchmark model
leads to only ∼ 24 events, to be compared to the 41.8 events ATLAS can exclude in this
bin.
We have checked that Zh associated production does not yield a sizable contribution to
the CMS and ATLAS multilepton analyses. In particular, these events dominantly populate
the CMS 4` bin with 75 GeV < M``1 < 105 GeV and M``2 < 75 GeV [367], in which the
signal would only be ∼ 0.2 events.
The inclusive multilepton search strategy pursued by CMS does a reasonable job of
constraining multileptonic Higgs decays when the mass splitting in the cascade decay is
sufficiently large that all four leptons can be identified at a reasonable rate. However the
rapid degradation of these limits as the mass splitting is squeezed suggests that further
adapting multilepton searches to the kinematics of exotic Higgs decays would be beneficial
in order to recover sensitivity to cascade decays with smaller mass splittings.
As the mass splitting is decreased, VBF and Wh associated production become more
important relative to gluon fusion. Although VBF production yields slightly higher-pT final
states than either gluon fusion or Wh, the Higgs exotic decay is still a lower-pT signal than
most BSM signals sought in multi-lepton searches. An analysis more tailored to the specific
kinematics of a 125 GeV Higgs could improve the reach. Imposing cuts on the transverse
mass of the leptons and the E/T could efficiently separate the Higgs signals from top and
fake backgrounds, so long as VBF is more important than Wh; it may also be beneficial to
target VBF production directly, by requiring the presence of tagging jets. In the CMS multi-
lepton searches, regardless of the mass splittings in the cascade, Wh production dominantly
populates the bin with three identified leptons, one OSSF pair with invariant mass below
the Z window, and zero τs and b-jets, in the lowest ST (HT ) bin. This is the same bin
that receives the greatest single contribution from gluon fusion as well. The background
composition in this bin contains a larger proportional contribution from fake leptons than in
bins with higher ST [297], suggesting tighter lepton ID may be beneficial in optimizing search
strategies for the relatively low-pT Higgs signal, as well as more aggressive b-jet rejection to
suppress backgrounds from top pair production. Further, ST regions designed for SM Higgs
production mechanisms could help by concentrating the VBF signal in a single bin (as gluon
fusion and Wh already are).
Finally, we comment on the case where the leptons form resonant pairs. In particular
134
let us consider the decay chains h → KvKv → 2ZD2πv → 4` + E/T , so that Br(h → 4` +
E/T )/Br(h→ BSM) = Br(ZD → `+`−)2. In general, the signal acceptance in the above multi-
lepton searches does not change substantially relative to the nonresonant signals. However,
the presence of the leptonic resonances makes these decays much easier to constrain. Once
again, limits will be highly sensitive to the BSM mass spectrum, which controls the lepton
pT s. In spectra giving rise to decays with little to no E/T , exclusions on the parent exotic
decay could approach the . 10−3 level obtained for h → 4` decays with no E/T (see §11),
with the sensitivity dropping rapidly as the spectrum is squeezed and the lepton acceptance
drops.
15. h→ 2` + E/T
In this section, we study exotic Higgs decays to final states that contain two isolated
leptons and missing energy, where the leptons do not reconstruct a resonance (we also
comment briefly on the case where they do). Models which realize these decays often also
realize decays with 4 leptons and missing energy, covered in §14.
15.1. Theoretical Motivation
In §14, we outlined many classes of theories where an initial decay h → XX is followed
by the decay X → ``E/T . One example, which produces an OSSF lepton pair, is the decay
of a neutralino χ2 through an off-shell Z boson to ``χ1. Similarly, a hidden sector meson
Kv could decay through an off-shell dark vector boson ZD into OSSF leptons plus a lighter,
detector-stable hidden meson, ``πv.
Decays where h→ 2`+E/T +X can arise in these theories in two ways. First, in a decay
that begins via h → χ2χ2, one of the χ2’s can decay to 2` + E/T while the other decays
to 2j + E/T or 2ν + E/T . Second, the Higgs will frequently also have the off-diagonal decay
h → χ2χ1, giving h → 2` + E/T . All of these decay chains result in an OSSF lepton pair
together with missing energy and potentially extra soft jets [371].
Another realization of the signature h→ 2`+ E/T is found in theories with a light sterile
neutrino, where the coupling yiNHLi gives rise to the decays h → νN , followed by both
N → `iW(∗) → `i`jν and N → νZ(∗) → ν`` [54, 371]. Decays through the (virtual)
135
W could yield opposite-sign dileptons with no flavor correlation, unlike the OSSF pair of
leptons generated through Z(∗) and Z(∗)D . These Higgs decays would also be accompanied
by Drell-Yan production of Nν, which yields a non-resonant contribution to the same final
states.
As discussed in §14, if there is a light bosonic state, the decay χ2 → χ1`` can proceed
via an intermediate on-shell state, χ2 → ZDχ1, aχ1, sχ1, such that the leptons reconstruct
a resonance. For dark vector bosons, the branching ratio to light leptons is appreciable for
any mZD < mh/2. For (pseudo-)scalars with mass-weighted couplings, such as can appear
in the the PQ-NMSSM [53], we need m(a,s) . 2mτ for muonic branching fractions to be
significant. This does not necessarily imply that the muons will be collimated, as the a(s)
is coming from a cascade decay, and depending on the particular values of m2, m1, may be
produced at relatively low pT . Nevertheless the experimental searches for high-pT isolated
leptons almost invariably require m`` > (10-12) GeV for all OSSF pairs in order to suppress
quarkonia backgrounds, making such searches insensitive to light bosons regardless of their
pT . We discuss the case of h→ (``) + E/T through a low-mass boson like a or s in §16.
Finally, we also comment that flavor-violating decays h → χχ followed by χ → `qq′
yield two leptons plus additional soft jets, albeit no missing energy. These decays can arise
from Higgs decay to neutralinos, which decay through R-parity violating operators such
as LiQjdk. They also occur in models where the Higgs decays to two heavy right-handed
neutrinos, followed by N → W (∗)` → qq′` [129]. Similar final states can arise in scotogenic
models [362, 363]. When the neutrino or neutralino is Majorana, the leptons may have the
same sign, yielding a distinctive signature.
15.2. Existing Experimental Searches and Limits
The signature of ≥ 2 leptons together with missing energy occurs in the SM decays of a
125 GeV Higgs boson: the decays of a Higgs into WW ∗, ττ and ZZ∗, with subsequent decays
of W/Z bosons and taus into leptons and neutrinos give rise to this final state. While the
decay h→ Z(∗)Z∗ → ``+νν suffers from a disadvantageous signal-to-background ratio, both
h→ WW ∗ → 2`+E/T and h→ ττ → 2`+E/T are standard SM Higgs search channels. These
SM leptons + invisible Higgs decays can, depending on kinematics, present an important
background for BSM Higgs searches in leptons plus missing energy final states. Conversely,
136
existing SM Higgs searches have sensitivity to begin to constrain BSM leptons + invisible
Higgs decays, though the tailoring of SM Higgs searches to SM decay kinematics reduces
their reach for BSM multi-lepton + missing energy decays [372]. Associated Wh production
also yields three-lepton final states, but at rates too small to be constrained by both ATLAS
and CMS multilepton searches [296, 297, 365, 366].
We will estimate the limits on a benchmark decay chain that begins with the off-diagonal
decay h→ χ1χ2, followed by χ2 → χ1 + 2` through an off-shell Z,
h→ χ1χ2 → 2`+ 2χ1. (99)
We will show results for the optimistic reference working point presented in the previous
section, where mχ1 = 20 GeV, mχ2 = 55 GeV. Limits for h→ χ2χ2 → 2`+E/T +X cascade
decays will be less constraining than those for the off-diagonal decay due to the reduced E/T .
For the decay h→ χ2χ1, depending on the masses m2,m1, the kinematics of the daughter
leptons and E/T are often broadly similar to the SM h → WW ∗ decay. Recalling that
Br(h → WW ∗ → 2`2ν) ≈ 0.26 × 0.103 and that Br(Z → ``) ≈ 0.102 (we include τs),
a Higgs with 10% branching fraction to χ1χ2 contributes roughly 40% the rate of the SM
WW ∗ dileptonic decay mode before acceptance is taken into account.
Performing a careful recast of SM h → WW ∗ searches is challenging as the sensitivity
to exotic signals is not straightforward to extract from the published experimental analy-
ses. CMS’ SM searches use multivariate discriminants to separate signal from background,
rendering a careful recast challenging except in the earliest analyses (such as [373]), which
are not constraining. Meanwhile, ATLAS’s full 7+8 TeV results [374] extract the SM signal
using a multichannel likelihood, and a recast would require use of the full likelihood func-
tion. Here our main aim is to estimate the BSM branching fraction into dileptonic modes,
which is allowed by SM Higgs searches. To this end we approximate the BSM acceptance
to be equal to the SM acceptance in the multivariate discriminants. This is a conservative
choice, but likely to be the correct order of magnitude for the particular benchmark model
we consider. For more general choices of m1, m2, the acceptance will often be significantly
reduced relative to this benchmark, as the daughter leptons may be much softer.
As in the previous section, to obtain these limits we use MadGraph 5 and Pythia 6 to
generate gluon fusion Higgs signal events, matched out to one jet. For CMS searches, we
employ a version of PGS tuned to CMS’ operating parameters. For ATLAS searches, we
137
60 80 100 120 140 160 180 2000.0
0.1
0.2
0.3
0.4
0.5
mTH2l, METLqHGeVL
1 N
dNk
dm
FIG. 34: Unit-normalized distributions of mT (2`, E/T ). The blue dashed line shows the ATLAS
prediction for SM h → WW ∗ events passing all selection criteria in both 7 and 8 TeV data
sets [374]. The purple dotted line shows the distribution for the BSM h→ 2`+ E/T events arising
from h→ χ2χ1 at the 8 TeV LHC in the benchmark model described in the text.
use parameterized lepton efficiencies as reported in the searches under consideration, with
jet clustering performed in FastJet. We neglect VBF production, as well as the VBF-like
event categories in the ATLAS and CMS searches.
The “cut-based” analysis of the full 7+8 TeV CMS 0j and 1j h → WW ∗ analysis [375]
employs a multivariate discriminant in states with same-flavor leptons to separate h→ WW ∗
signal from Drell-Yan pair production. Approximating the efficiency of this multivariate
discriminant at the SM Higgs-like value ε ≈ 0.5 on the BSM decay mode h→ χ2χ1 → 2`+E/T ,
and combining the effect of this multivariate cut with the rest of the analysis selection, we
can estimate the ratio of the BSM signal to the SM signal. Using CMS’ best fit for the SM
signal strength µ in the h→ WW ∗ mode in the 0 and 1 jet categories,
µ|fit = 0.79± 0.38, (100)
we estimateσ(pp→ h)
σ(pp→ h)|SMBr(h→ χ1χ2) . 1.0 (101)
for the reference benchmark point. Again, this limit includes an assumed factor of Br(Z →
`+`−) ∼ 0.102; decay chains with off-shell dark photons, which have leptonic branching
fractions roughly 4 times larger, are subject to the tighter constraint Br(h→ χ2χ1) . 0.24.
Meanwhile, in the ATLAS analysis [374], the final step in the analysis is fitting SM
signal and background distributions in the transverse mass variable mT (2`, E/T ). ATLAS’
138
background-subtracted predictions for the SM signal strength are shown in Fig. 34, together
with the prediction from the BSM benchmark, to indicate the degree of similarity between
the two signals in the final discriminating variable. The cuts employed in the ATLAS
analysis give comparatively less sensitivity to the BSM signal than do the CMS cuts. As a
consequence, under the simplifying assumption that the SM and BSM signals are extracted
with similar efficiency in the final fit, no limit is placed on the branching fraction into the
BSM final state.
Since the signal investigated in this section contributes almost entirely to same-flavor
final states, better sensitivity could be obtained by considering different-flavor and same-
flavor final states separately. As our recasting is highly approximate due to the lack of
information about the multivariate discriminants employed in the same-flavor final states,
we will simply mention this as one obvious avenue for improving on the approximate bound
shown in Eq. (101). In cases where the two leptons reconstruct a resonance, significantly
better limits may be possible. Meanwhile the heavy neutrino decay through a (virtual)
W , which does contribute to different-flavor final states, would show interesting departures
from flavor universality depending on the flavor mixings in the neutrino sector; this heavy
neutrino model should be looked for simultaneously in Drell-Yan and Higgs decays as the
ratio of the two signals is fixed.
16. h→ ONE LEPTON-JET + X
In this and the following section, we study exotic Higgs decays to final states that contain
one or two low-mass resonant lepton pairs. Higgs decays to collimated pairs of leptons (here
` = e, µ but not τ) have been a focus of much experimental and theoretical work. Searches
for collimated pairs of leptons are typically carried out inclusively, that is, no attempt to
reconstruct the Higgs mass is made. Thus the same searches constrain decays both with
and without the presence of E/T , although events with E/T (or other Higgs daughter products,
such as soft jets) will typically have reduced acceptance. In this section, we consider Higgs
decays to one lepton-jet+X, and in the following section we consider Higgs decays to two
lepton-jets+X. For simplicity we focus on simple lepton-jets, consisting of a collimated pair
of either electrons or muons; complex lepton-jets, which have a larger and more variable
particle content that can involve hadrons and detector-stable states as well as leptons, are
139
important and interesting signals, but less transparent to survey.
In the current section we study Higgs decays to one (simple) lepton-jet+X. Because
experimental backgrounds for a single lepton-jet are higher than those for two, traditionally
the focus has been on signals with two lepton-jets. In this section we emphasize, firstly, that
there are well-motivated signals that produce a single lepton-jet only or dominantly, and
secondly, that exclusive analyses targeting these states can yield meaningful sensitivity to
these decays.
The opening angle of two partons coming from a parent particle X can be roughly es-
timated as ∆R ' 2mX/pT,X . We can estimate pT,X ∼ 50 GeV, for a particle X coming
from the decay of a 125 GeV Higgs produced at rest. Partons from the X decay are then
typically separated by ∆R < 0.2 when mX . 5 GeV. Therefore, we expect to have a Higgs
decaying into collimated leptons that fail typical isolation cuts requiring ∆R > 0.4 if the
parent particle X has a mass of the order of 10 GeV or less. Meanwhile if the parent particle
X is produced in a cascade decay instead of directly, it will be less boosted. Clearly the
transition between having isolated leptons and collimated leptons happens smoothly as a
function of the parent particle mass mX . The reader may also be interested in §11, which
considers isolated leptons with m`` > 4 GeV.
16.1. Theoretical Motivation
One theory that realizes the decay h → (µµ) + E/T is is the PQ-symmetric limit of the
NMSSM [52, 53]. In this limit, the degrees of freedom (s, a, χ1) (scalar, pseudo-scalar, and
fermion, respectively) comprising the singlet multiplet are all light. Decays of the Higgs
to h → χ1χ2 or h → χ2χ2, with subsequent decays χ2 → χ1s, χ1a, give Higgs decay
signatures with missing energy in the final state. In an appreciable portion of parameter
space, these decays can dominate the exotic Higgs branching fraction, as detailed in §1.3.8
and Refs. [52, 53]. If s (or a) is very light, with mass order m . O(1) GeV or below,
phase space forbids decays to heavier fermions and the branching fraction into light leptons
becomes appreciable (O(10%); see, e.g., Fig. 7). The resulting signatures are dileptons +
E/T for h → χ1χ2 and four leptons + E/T for h → χ2χ2, which correspond to the type-I and
type-III decay topologies presented in §14. The s(a) is produced with a pT that is dependent
on the masses of χ2 and χ1, but in the regime where decays to muons dominate, typically
140
we will have pT (a,s) � ma,s, and the daughter muons will be collimated: ∆R`` . 0.1.
Dark vector boson models can also realize the collimated leptons+E/T Higgs decay signa-
ture. In a supersymmetric context, χ2 would now be mainly bino and χ1 a dark photino, but
in this case the off-diagonal h → χ2χ1 decay can only be important if the decay h → χ2χ2
is kinematically forbidden. In a more general hidden sector, the role of the neutralinos χi
may be played instead by hidden sector mesons Kv, πv or similar states, see §1.3.10. Dark
photon models can also yield Higgs decays of type II topology (see §14). In this case, the
Higgs decays directly to dark vectors, h→ ZDZD, followed by ZD → lepton-jet on one side
and ZD → invisible on the other. Here the invisible states are detector-stable hidden sector
states, perhaps dark photinos [51, 148, 361, 376]; the relative branching fractions to leptons,
E/T , and other SM partons are model-dependent. Similar signatures can be obtained in the
R-symmetric NMSSM if the light pseudo-scalar is coupled to a hidden sector. Another pos-
sible realization of the type II topology is provided by the decay h → ZZD, followed by
Z → νν.
Also a possibility are decays h → (µµ) + (jj), i.e., where the lepton-jet recoils against
hadronic activity. This kind of decay arises in, e.g., the R-symmetric limit of the NMSSM,
where h→ aa is followed by a→ µµ on one side of the event, and a→ hadrons on the other.
As Br(a → µµ) . 0.1 even below the τ threshold, Br(h → (µµ)(jj)) > Br(h → 2(µµ));
however the 2(µµ) final state has notably lower background, as well as sharper resolution.
Similarly, h → ZDZD → (``)(jj) leads to a lepton-jet balanced against a “weird” hadronic
jet.
Unlike the NMSSM (pseudo)scalars, dark photons have appreciable branching fractions
to light leptons even for large masses mZD . However, possible connections with cosmic
ray anomalies [135, 136] and the discrepancy between the measured and calculated muon
anomalous magnetic moment [139] have stimulated interest in dark vectors with a mass at
or below the GeV scale, thus involving collimated leptons in the final state. For discussion
of dark vectors outside the collimated regime, see §10 and §11.
16.2. Existing Collider Studies
A dedicated analysis for h → χ1χ2 → `+`− + E/T is presented in [53], which indicates
that the 8 TeV LHC could have good sensitivity to this final state when a targeted search
141
ms mh mχ1 mχ2
1 GeV 125 GeV 10 GeV 80 GeV
TABLE IX: Mass parameters of the h→ collimated leptons + E/T benchmark model.
is performed that exploits the E/T in the final state from the Higgs decay. As an illustration,
the analysis focuses on a benchmark inspired by the PQ-symmetric limit of the NMSSM,
with a light scalar(pseudoscalar) resonance s(a) set to have a mass of 1 GeV (see Table IX).
The analysis focuses on the W±h production mode where the W decays leptonically.
The resulting signature contains one hard lepton (e, µ) from the W decay, two collimated
muons, and E/T . Since there are no jets in the hard scattering process, the W+jets, Z+jets,
and tt backgrounds can be efficiently eliminated with a jet veto. The diboson WZ and
ZZ backgrounds are be removed by a dimuon mass window cut. A muon isolation cut is
applied to remove the low-mass dimuon background from meson decays, which requires the
transverse momentum sum of hadronic jets (excluding the contribution from any nearby
muons) in a cone of R = 0.4 around each muon candidate to be less than 5 GeV. Then the
light resonance can be reconstructed via the two nearby muons, and the main background
is Wγ∗/Z, with γ∗/Z decaying into µ+µ−. A trilepton trigger is assumed in the analysis,
though alternatively, one can trigger on the single lepton from the W decay. The analysis
indicates that, with 20 fb−1 data, a sensitivity S/√B > 6σ can be achieved at the 8 TeV